Hello Takashi,
-----Original Message----- From: Takashi Iwai [mailto:tiwai@suse.de] Sent: Wednesday, November 20, 2019 9:49 PM To: Gabbasov, Andrew Cc: alsa-devel@alsa-project.org; linux-kernel@vger.kernel.org; Jaroslav Kysela; Takashi Iwai; Timo Wischer Subject: Re: [PATCH v5 0/7] ALSA: aloop: Support sound timer as clock source instead of jiffies
On Wed, 20 Nov 2019 18:49:48 +0100, Andrew Gabbasov wrote:
This patch set is an updated version of patches by Timo Wischer: https://mailman.alsa-project.org/pipermail/alsa-devel/2019-
March/146871.html
This patch set is required for forwarding audio data between a HW sound card and an aloop device without the usage of an asynchronous sample
rate
converter.
Most of sound and timers related code is kept the same as in previous
set.
The code, related to snd_pcm_link() functionality and its using for timer source setting, is removed (as rejected earlier). The changes in
this
update are mainly related to the parameters handling and some cleanup.
The timer source can be initially selected by "timer_source" kernel
module
parameter. It is supposed to have the following format: [<pref>:](<card name>|<card idx>)[{.,}<dev idx>[{.,}<subdev idx>]] For example: "hw:I82801AAICH.1.0", or "1.1", or just "I82801AAICH". (Prefix is ignored, just allowed here to be able to use the strings, the user got used to). Although the parsing function recognizes both '.' and ',' as a
separator,
module parameters handling routines use ',' to separate parameters for different module instances (array elements), so we have to use '.' to separate device and subdevice numbers from the card name or number in module parameters. Empty string indicates using jiffies as a timer source.
Besides "static" selection of timer source at module load time, it is possible to dynamically change it via sound "info" interface (using "/proc/asound/<card>/timer_source" file in read-write mode. The contents of this file is used as a timer source string for a particular loopback card, e.g. "hw:0,0,0" (and here ',' can be used as a separator).
The timer source string value can be changed at any time, but it is latched by PCM substream open callback "loopback_open()" (the first one for a particular cable). At this point it is actually used, that is the string is parsed, and the timer is looked up and opened. This seems to be a good trade-off between flexibility of updates and synchronizations or racing complexity.
The timer source is set for a loopback card (the same as initial setting by module parameter), but every cable uses the value, current at the
moment
of opening. Theoretically, it's possible to set the timer source for
each
cable independently (via separate files), but it would be inconsistent with the initial setting via module parameters on a per-card basis.
v2: https://mailman.alsa-project.org/pipermail/alsa-devel/2019-
November/157961.html
v3: https://mailman.alsa-project.org/pipermail/alsa-devel/2019-
November/158312.html
- Change sound card lookup to use snd_card_ref() and avoid direct access to sound cards array
- Squash commits on returning error codes for timer start and stop
- Some locking related fixes
- Some code cleanup
v4: https://mailman.alsa-project.org/pipermail/alsa-devel/2019-
November/158896.html
- Change to use updated API for snd_timer_open() (separate timer
instance)
- Change to use snd_timer_close() returning void
- Some code cleanup
v5:
- Change to initialize timer event tasklet before calling
snd_timer_open()
Andrew Gabbasov (1): ALSA: aloop: Support runtime change of snd_timer via info interface
Timo Wischer (6): ALSA: aloop: Describe units of variables ALSA: aloop: Support return of error code for timer start and stop ALSA: aloop: Use callback functions for timer specific implementations ALSA: aloop: Rename all jiffies timer specific functions ALSA: aloop: Move CABLE_VALID_BOTH to the top of file ALSA: aloop: Support selection of snd_timer instead of jiffies
Now applied all patches. Thanks.
Thank you very much!
Unfortunately, I found some issue with locking in patch 6 (sorry for late finding): loopback_parse_timer_id() uses snd_card_ref(), that can lock on mutex, also snd_timer_instance_new() uses non-atomic allocation, that can sleep. So, both functions can not be called from loopback_snd_timer_open() with cable->lock spinlock locked.
Also, most part of loopback_snd_timer_open() function body works when the opposite stream of the same cable does not yet exist, and the current stream is not yet completely open and can't be running, so existing locking of loopback->cable_lock mutex is enough to protect from conflicts with simultaneous opening or closing.
So, it looks like locking of cable->lock spinlock is not needed in loopback_snd_timer_open() at all (similarly to loopback_snd_timer_close_cable()).
I'm sending a separate patch to fix this: https://mailman.alsa-project.org/pipermail/alsa-devel/2019-November/159087.h tml
Also, when you requested to move tasklet_init() call to before snd_timer_open() call, I should have guessed to look into the reverse closing side too ;-) Now, I have to add one more patch to move tasklet_kill() call to after snd_timer_close(). https://mailman.alsa-project.org/pipermail/alsa-devel/2019-November/159088.h tml
Please, take a look at the patches.
Thanks!
Best regards, Andrew