2005-01-06 01:43:34 +00:00
|
|
|
/*-
|
2003-09-07 16:28:03 +00:00
|
|
|
* Copyright (c) 1999 Cameron Grant <cg@freebsd.org>
|
1999-09-01 04:08:39 +00:00
|
|
|
* Portions Copyright by Luigi Rizzo - 1997-99
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
* are met:
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
2005-06-10 21:33:14 +00:00
|
|
|
#include "opt_isa.h"
|
|
|
|
|
1999-11-20 16:50:33 +00:00
|
|
|
#include <dev/sound/pcm/sound.h>
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2000-12-18 01:36:41 +00:00
|
|
|
#include "feeder_if.h"
|
|
|
|
|
2001-08-23 11:30:52 +00:00
|
|
|
SND_DECLARE_FILE("$FreeBSD$");
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
#define MIN_CHUNK_SIZE 256 /* for uiomove etc. */
|
2006-01-24 01:10:07 +00:00
|
|
|
#if 0
|
1999-09-01 04:08:39 +00:00
|
|
|
#define DMA_ALIGN_THRESHOLD 4
|
|
|
|
#define DMA_ALIGN_MASK (~(DMA_ALIGN_THRESHOLD - 1))
|
2006-01-24 01:10:07 +00:00
|
|
|
#endif
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
#define CANCHANGE(c) (!(c->flags & CHN_F_TRIGGERED))
|
2000-11-07 00:32:35 +00:00
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
#define BUF_PARENT(c, b) \
|
|
|
|
(((c) != NULL && (c)->parentchannel != NULL && \
|
|
|
|
(c)->parentchannel->bufhard != NULL) ? \
|
|
|
|
(c)->parentchannel->bufhard : (b))
|
|
|
|
|
1999-12-29 03:46:54 +00:00
|
|
|
/*
|
|
|
|
#define DEB(x) x
|
|
|
|
*/
|
2000-11-07 00:32:35 +00:00
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
int report_soft_formats = 1;
|
|
|
|
SYSCTL_INT(_hw_snd, OID_AUTO, report_soft_formats, CTLFLAG_RW,
|
|
|
|
&report_soft_formats, 1, "report software-emulated formats");
|
|
|
|
|
|
|
|
int chn_latency = CHN_LATENCY_DEFAULT;
|
|
|
|
TUNABLE_INT("hw.snd.latency", &chn_latency);
|
2001-09-18 14:45:09 +00:00
|
|
|
|
|
|
|
static int
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
sysctl_hw_snd_latency(SYSCTL_HANDLER_ARGS)
|
2001-09-18 14:45:09 +00:00
|
|
|
{
|
|
|
|
int err, val;
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
val = chn_latency;
|
2001-09-18 14:45:09 +00:00
|
|
|
err = sysctl_handle_int(oidp, &val, sizeof(val), req);
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (val < CHN_LATENCY_MIN || val > CHN_LATENCY_MAX)
|
2001-09-18 14:45:09 +00:00
|
|
|
err = EINVAL;
|
|
|
|
else
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
chn_latency = val;
|
2001-09-18 14:45:09 +00:00
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
SYSCTL_PROC(_hw_snd, OID_AUTO, latency, CTLTYPE_INT | CTLFLAG_RW,
|
|
|
|
0, sizeof(int), sysctl_hw_snd_latency, "I",
|
|
|
|
"buffering latency (0=low ... 10=high)");
|
|
|
|
|
|
|
|
int chn_latency_profile = CHN_LATENCY_PROFILE_DEFAULT;
|
|
|
|
TUNABLE_INT("hw.snd.latency_profile", &chn_latency_profile);
|
|
|
|
|
|
|
|
static int
|
|
|
|
sysctl_hw_snd_latency_profile(SYSCTL_HANDLER_ARGS)
|
|
|
|
{
|
|
|
|
int err, val;
|
|
|
|
|
|
|
|
val = chn_latency_profile;
|
|
|
|
err = sysctl_handle_int(oidp, &val, sizeof(val), req);
|
|
|
|
if (val < CHN_LATENCY_PROFILE_MIN || val > CHN_LATENCY_PROFILE_MAX)
|
|
|
|
err = EINVAL;
|
|
|
|
else
|
|
|
|
chn_latency_profile = val;
|
|
|
|
|
|
|
|
return err;
|
|
|
|
}
|
|
|
|
SYSCTL_PROC(_hw_snd, OID_AUTO, latency_profile, CTLTYPE_INT | CTLFLAG_RW,
|
|
|
|
0, sizeof(int), sysctl_hw_snd_latency_profile, "I",
|
|
|
|
"buffering latency profile (0=aggresive 1=safe)");
|
2001-09-18 14:45:09 +00:00
|
|
|
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
/**
|
|
|
|
* @brief Channel sync group lock
|
|
|
|
*
|
|
|
|
* Clients should acquire this lock @b without holding any channel locks
|
|
|
|
* before touching syncgroups or the main syncgroup list.
|
|
|
|
*/
|
|
|
|
struct mtx snd_pcm_syncgroups_mtx;
|
|
|
|
MTX_SYSINIT(pcm_syncgroup, &snd_pcm_syncgroups_mtx, "PCM channel sync group lock", MTX_DEF);
|
|
|
|
/**
|
|
|
|
* @brief syncgroups' master list
|
|
|
|
*
|
|
|
|
* Each time a channel syncgroup is created, it's added to this list. This
|
|
|
|
* list should only be accessed with @sa snd_pcm_syncgroups_mtx held.
|
|
|
|
*
|
|
|
|
* See SNDCTL_DSP_SYNCGROUP for more information.
|
|
|
|
*/
|
|
|
|
struct pcm_synclist snd_pcm_syncgroups = SLIST_HEAD_INITIALIZER(head);
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
static int chn_buildfeeder(struct pcm_channel *c);
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
static void
|
2003-12-05 02:08:13 +00:00
|
|
|
chn_lockinit(struct pcm_channel *c, int dir)
|
2001-03-24 23:10:29 +00:00
|
|
|
{
|
2004-02-28 19:47:02 +00:00
|
|
|
switch(dir) {
|
|
|
|
case PCMDIR_PLAY:
|
|
|
|
c->lock = snd_mtxcreate(c->name, "pcm play channel");
|
|
|
|
break;
|
|
|
|
case PCMDIR_REC:
|
|
|
|
c->lock = snd_mtxcreate(c->name, "pcm record channel");
|
|
|
|
break;
|
|
|
|
case PCMDIR_VIRTUAL:
|
|
|
|
c->lock = snd_mtxcreate(c->name, "pcm virtual play channel");
|
|
|
|
break;
|
|
|
|
case 0:
|
|
|
|
c->lock = snd_mtxcreate(c->name, "pcm fake channel");
|
|
|
|
break;
|
|
|
|
}
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
|
|
|
|
cv_init(&c->cv, c->name);
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
static void
|
|
|
|
chn_lockdestroy(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
snd_mtxfree(c->lock);
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
cv_destroy(&c->cv);
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
2001-02-07 20:45:40 +00:00
|
|
|
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
/**
|
|
|
|
* @brief Determine channel is ready for I/O
|
|
|
|
*
|
|
|
|
* @retval 1 = ready for I/O
|
|
|
|
* @retval 0 = not ready for I/O
|
|
|
|
*/
|
1999-09-01 04:08:39 +00:00
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_polltrigger(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
unsigned amt, lim;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
if (c->flags & CHN_F_MAPPED) {
|
|
|
|
if (sndbuf_getprevblocks(bs) == 0)
|
|
|
|
return 1;
|
2001-03-25 21:43:24 +00:00
|
|
|
else
|
2001-03-24 23:10:29 +00:00
|
|
|
return (sndbuf_getblocks(bs) > sndbuf_getprevblocks(bs))? 1 : 0;
|
|
|
|
} else {
|
|
|
|
amt = (c->direction == PCMDIR_PLAY)? sndbuf_getfree(bs) : sndbuf_getready(bs);
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#if 0
|
2001-03-24 23:10:29 +00:00
|
|
|
lim = (c->flags & CHN_F_HAS_SIZE)? sndbuf_getblksz(bs) : 1;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#endif
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
lim = c->lw;
|
|
|
|
return (amt >= lim) ? 1 : 0;
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
|
|
|
return 0;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_pollreset(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
sndbuf_updateprevtotal(bs);
|
1999-09-01 04:08:39 +00:00
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_wakeup(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2004-02-19 01:07:15 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
struct pcmchan_children *pce;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2004-02-19 01:07:15 +00:00
|
|
|
if (SLIST_EMPTY(&c->children)) {
|
|
|
|
if (SEL_WAITING(sndbuf_getsel(bs)) && chn_polltrigger(c))
|
2004-02-20 01:24:57 +00:00
|
|
|
selwakeuppri(sndbuf_getsel(bs), PRIBIO);
|
2004-02-19 01:07:15 +00:00
|
|
|
} else {
|
|
|
|
SLIST_FOREACH(pce, &c->children, link) {
|
|
|
|
CHN_LOCK(pce->channel);
|
|
|
|
chn_wakeup(pce->channel);
|
|
|
|
CHN_UNLOCK(pce->channel);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
wakeup_one(bs);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
static int
|
|
|
|
chn_sleep(struct pcm_channel *c, char *str, int timeout)
|
1999-12-13 03:29:09 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
2001-06-23 17:36:51 +00:00
|
|
|
#ifdef USING_MUTEX
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = msleep(bs, c->lock, PRIBIO | PCATCH, str, timeout);
|
2001-06-23 17:36:51 +00:00
|
|
|
#else
|
|
|
|
ret = tsleep(bs, PRIBIO | PCATCH, str, timeout);
|
|
|
|
#endif
|
2001-02-07 06:48:28 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
return ret;
|
1999-12-13 03:29:09 +00:00
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
/*
|
|
|
|
* chn_dmaupdate() tracks the status of a dma transfer,
|
2005-10-02 15:37:40 +00:00
|
|
|
* updating pointers.
|
1999-09-01 04:08:39 +00:00
|
|
|
*/
|
1999-12-05 19:09:13 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
static unsigned int
|
|
|
|
chn_dmaupdate(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
unsigned int delta, old, hwptr, amt;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-05-27 17:22:00 +00:00
|
|
|
KASSERT(sndbuf_getsize(b) > 0, ("bufsize == 0"));
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2001-12-21 22:34:01 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
old = sndbuf_gethwptr(b);
|
1999-12-05 19:09:13 +00:00
|
|
|
hwptr = chn_getptr(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
delta = (sndbuf_getsize(b) + hwptr - old) % sndbuf_getsize(b);
|
|
|
|
sndbuf_sethwptr(b, hwptr);
|
|
|
|
|
2001-02-07 06:48:28 +00:00
|
|
|
DEB(
|
2001-03-24 23:10:29 +00:00
|
|
|
if (delta >= ((sndbuf_getsize(b) * 15) / 16)) {
|
2000-08-09 00:42:00 +00:00
|
|
|
if (!(c->flags & (CHN_F_CLOSING | CHN_F_ABORTING)))
|
2002-01-25 04:14:12 +00:00
|
|
|
device_printf(c->dev, "hwptr went backwards %d -> %d\n", old, hwptr);
|
2000-08-09 00:42:00 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
);
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
if (c->direction == PCMDIR_PLAY) {
|
|
|
|
amt = MIN(delta, sndbuf_getready(b));
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
amt -= amt % sndbuf_getbps(b);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (amt > 0)
|
|
|
|
sndbuf_dispose(b, NULL, amt);
|
1999-09-01 04:08:39 +00:00
|
|
|
} else {
|
2001-03-24 23:10:29 +00:00
|
|
|
amt = MIN(delta, sndbuf_getfree(b));
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
amt -= amt % sndbuf_getbps(b);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (amt > 0)
|
|
|
|
sndbuf_acquire(b, NULL, amt);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (snd_verbose > 2 && (c->flags & CHN_F_TRIGGERED) && delta == 0) {
|
|
|
|
device_printf(c->dev, "WARNING: PCMDIR_%s DMA completion "
|
|
|
|
"too fast/slow ! hwptr=%u, old=%u "
|
|
|
|
"delta=%u amt=%u ready=%u free=%u\n",
|
|
|
|
(c->direction == PCMDIR_PLAY) ? "PLAY" : "REC",
|
|
|
|
hwptr, old, delta, amt,
|
|
|
|
sndbuf_getready(b), sndbuf_getfree(b));
|
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
|
|
|
|
return delta;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
1999-12-29 03:46:54 +00:00
|
|
|
void
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_wrupdate(struct pcm_channel *c)
|
1999-12-05 19:09:13 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
int ret;
|
1999-12-05 19:09:13 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
KASSERT(c->direction == PCMDIR_PLAY, ("chn_wrupdate on bad channel"));
|
|
|
|
|
2001-06-07 20:06:22 +00:00
|
|
|
if ((c->flags & (CHN_F_MAPPED | CHN_F_VIRTUAL)) || !(c->flags & CHN_F_TRIGGERED))
|
2001-03-24 23:10:29 +00:00
|
|
|
return;
|
|
|
|
chn_dmaupdate(c);
|
|
|
|
ret = chn_wrfeed(c);
|
|
|
|
/* tell the driver we've updated the primary buffer */
|
|
|
|
chn_trigger(c, PCMTRIG_EMLDMAWR);
|
|
|
|
DEB(if (ret)
|
|
|
|
printf("chn_wrupdate: chn_wrfeed returned %d\n", ret);)
|
1999-12-13 03:29:09 +00:00
|
|
|
|
2001-02-07 20:45:40 +00:00
|
|
|
}
|
1999-12-13 03:29:09 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
int
|
|
|
|
chn_wrfeed(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
unsigned int ret, amt;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#if 0
|
|
|
|
DEB(
|
2001-03-24 23:10:29 +00:00
|
|
|
if (c->flags & CHN_F_CLOSING) {
|
|
|
|
sndbuf_dump(b, "b", 0x02);
|
|
|
|
sndbuf_dump(bs, "bs", 0x02);
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
})
|
|
|
|
#endif
|
2001-03-24 23:10:29 +00:00
|
|
|
|
2001-12-21 22:34:01 +00:00
|
|
|
if (c->flags & CHN_F_MAPPED)
|
|
|
|
sndbuf_acquire(bs, NULL, sndbuf_getfree(bs));
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
amt = sndbuf_getfree(b);
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
DEB(if (amt > sndbuf_getsize(bs) &&
|
|
|
|
sndbuf_getbps(bs) >= sndbuf_getbps(b)) {
|
|
|
|
printf("%s(%s): amt %d > source size %d, flags 0x%x", __func__, c->name,
|
|
|
|
amt, sndbuf_getsize(bs), c->flags);
|
|
|
|
});
|
2001-12-21 22:34:01 +00:00
|
|
|
|
2005-12-30 07:33:01 +00:00
|
|
|
ret = (amt > 0) ? sndbuf_feed(bs, b, c, c->feeder, amt) : ENOSPC;
|
|
|
|
/*
|
|
|
|
* Possible xruns. There should be no empty space left in buffer.
|
|
|
|
*/
|
|
|
|
if (sndbuf_getfree(b) > 0)
|
|
|
|
c->xruns++;
|
Whats New:
1. Support wide range sampling rate, as low as 1hz up to int32 max
(which is, insane) through new feeder_rate, multiple precisions
choice (32/64 bit converter). This is indeed, quite insane, but it
does give us more room and flexibility. Plenty sysctl options to
adjust resampling characteristics.
2. Support 24/32 bit pcm format conversion through new, much improved,
simplified and optimized feeder_fmt.
Changes:
1. buffer.c / dsp.c / sound.h
* Support for 24/32 AFMT.
2. feeder_rate.c
* New implementation of sampling rate conversion with 32/64 bit
precision, 1 - int32max hz (which is, ridiculous, yet very
addictive). Much improved / smarter buffer management to not
cause any missing samples at the end of conversion process
* Tunable sysctls for various aspect:
hw.snd.feeder_rate_ratemin - minimum allowable sampling rate
(default to 4000)
hw.snd.feeder_rate_ratemax - maximum allowable sampling rate
(default to 1102500)
hw.snd.feeder_rate_buffersize - conversion buffer size
(default to 8192)
hw.snd.feeder_rate_scaling - scaling / conversion method
(please refer to the source for explaination). Default to
previous implementation type.
3. feeder_fmt.c / sound.h
* New implementation, support for 24/32bit conversion, optimized,
and simplified. Few routines has been removed (8 to xlaw, 16 to
8). It just doesn't make sense.
4. channel.c
* Support for 24/32 AFMT
* Fix wrong xruns increment, causing incorrect underruns statistic
while using vchans.
5. vchan.c
* Support for 24/32 AFMT
* Proper speed / rate detection especially for fixed rate ac97.
User can override it using kernel hint:
hint.pcm.<unit>.vchanrate="xxxx".
Notes / Issues:
* Virtual Channels (vchans)
Enabling vchans can really, really help to solve overrun
issues. This is quite understandable, because it operates
entirely within its own buffering system without relying on
hardware interrupt / state. Even if you don't need vchan,
just enable single channel can help much. Few soundcards
(notably via8233x, sblive, possibly others) have their own
hardware multi channel, and this is unfortunately beyond
vchan reachability.
* The arrival of 24/32 also come with a price. Applications
that can do 24/32bit playback need to be recompiled (notably
mplayer). Use (recompiled) mplayer to experiment / test /
debug this various format using -af format=fmt. Note that
24bit seeking in mplayer is a little bit broken, sometimes
can cause silence or loud static noise. Pausing / seeking
few times can solve this problem.
You don't have to rebuild world entirely for this. Simply
copy /usr/src/sys/sys/soundcard.h to
/usr/include/sys/soundcard.h would suffice. Few drivers also
need recompilation, and this can be done via
/usr/src/sys/modules/sound/.
Support for 24bit hardware playback is beyond the scope of
this changes. That would require spessific hardware driver
changes.
* Don't expect playing 9999999999hz is a wise decision. Be
reasonable. The new feeder_rate implemention provide
flexibility, not insanity. You can easily chew up your CPU
with this kind of mind instability. Please use proper
mosquito repellent device for this obvious cracked brain
attempt. As for testing purposes, you can use (again)
mplayer to generate / play with different sampling rate. Use
something like "mplayer -af resample=192000:0:0 <files>".
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@
2005-07-31 16:16:22 +00:00
|
|
|
|
2001-04-08 20:20:52 +00:00
|
|
|
if (ret == 0 && sndbuf_getfree(b) < amt)
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_wakeup(c);
|
2001-12-21 22:34:01 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
return ret;
|
1999-12-13 03:29:09 +00:00
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
static void
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_wrintr(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
int ret;
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
/* update pointers in primary buffer */
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_dmaupdate(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
/* ...and feed from secondary to primary */
|
|
|
|
ret = chn_wrfeed(c);
|
|
|
|
/* tell the driver we've updated the primary buffer */
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_trigger(c, PCMTRIG_EMLDMAWR);
|
2001-03-24 23:10:29 +00:00
|
|
|
DEB(if (ret)
|
|
|
|
printf("chn_wrintr: chn_wrfeed returned %d\n", ret);)
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2001-08-29 02:49:54 +00:00
|
|
|
* user write routine - uiomove data into secondary buffer, trigger if necessary
|
2001-03-24 23:10:29 +00:00
|
|
|
* if blocking, sleep, rinse and repeat.
|
1999-09-01 04:08:39 +00:00
|
|
|
*
|
2001-03-24 23:10:29 +00:00
|
|
|
* called externally, so must handle locking
|
1999-09-01 04:08:39 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_write(struct pcm_channel *c, struct uio *buf)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2006-12-07 07:48:58 +00:00
|
|
|
int ret, count, sz;
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
2003-11-27 19:51:44 +00:00
|
|
|
void *off;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
int t, x, togo, p;
|
1999-12-29 03:46:54 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
1999-12-29 03:46:54 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = 0;
|
2001-03-25 21:43:24 +00:00
|
|
|
count = hz;
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
while (!ret && (buf->uio_resid > 0) && (count > 0)) {
|
|
|
|
sz = sndbuf_getfree(bs);
|
|
|
|
if (sz == 0) {
|
|
|
|
if (c->flags & CHN_F_NBIO)
|
|
|
|
ret = EWOULDBLOCK;
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
else if (c->flags & CHN_F_NOTRIGGER) {
|
|
|
|
/**
|
|
|
|
* @todo Evaluate whether EAGAIN is truly desirable.
|
|
|
|
* 4Front drivers behave like this, but I'm
|
|
|
|
* not sure if it at all violates the "write
|
|
|
|
* should be allowed to block" model.
|
|
|
|
*
|
|
|
|
* The idea is that, while set with CHN_F_NOTRIGGER,
|
|
|
|
* a channel isn't playing, *but* without this we
|
|
|
|
* end up with "interrupt timeout / channel dead".
|
|
|
|
*/
|
|
|
|
ret = EAGAIN;
|
|
|
|
} else {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
ret = chn_sleep(c, "pcmwr", c->timeout);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (ret == EWOULDBLOCK) {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
count -= c->timeout;
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = 0;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
} else if (ret == ERESTART || ret == EINTR) {
|
|
|
|
c->flags |= CHN_F_ABORTING;
|
|
|
|
return ret;
|
2001-03-24 23:10:29 +00:00
|
|
|
} else if (ret == 0)
|
2001-03-25 21:43:24 +00:00
|
|
|
count = hz;
|
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
} else {
|
|
|
|
sz = MIN(sz, buf->uio_resid);
|
|
|
|
KASSERT(sz > 0, ("confusion in chn_write"));
|
|
|
|
/* printf("sz: %d\n", sz); */
|
2003-11-27 19:51:44 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* The following assumes that the free space in
|
|
|
|
* the buffer can never be less around the
|
|
|
|
* unlock-uiomove-lock sequence.
|
|
|
|
*/
|
|
|
|
togo = sz;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
while (ret == 0 && togo > 0) {
|
2003-11-27 19:51:44 +00:00
|
|
|
p = sndbuf_getfreeptr(bs);
|
|
|
|
t = MIN(togo, sndbuf_getsize(bs) - p);
|
|
|
|
off = sndbuf_getbufofs(bs, p);
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
ret = uiomove(off, t, buf);
|
|
|
|
CHN_LOCK(c);
|
|
|
|
togo -= t;
|
|
|
|
x = sndbuf_acquire(bs, NULL, t);
|
|
|
|
}
|
|
|
|
ret = 0;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (!(c->flags & CHN_F_TRIGGERED))
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_start(c, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/* printf("ret: %d left: %d\n", ret, buf->uio_resid); */
|
|
|
|
|
|
|
|
if (count <= 0) {
|
2001-03-25 21:43:24 +00:00
|
|
|
c->flags |= CHN_F_DEAD;
|
2001-09-05 16:28:41 +00:00
|
|
|
printf("%s: play interrupt timeout, channel dead\n", c->name);
|
2001-03-25 21:43:24 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
|
1999-12-29 03:46:54 +00:00
|
|
|
return ret;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
2006-01-01 18:16:13 +00:00
|
|
|
#if 0
|
2000-05-26 21:55:13 +00:00
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_rddump(struct pcm_channel *c, unsigned int cnt)
|
2000-05-26 21:55:13 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#if 0
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
static u_int32_t kk = 0;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
printf("%u: dumping %d bytes\n", ++kk, cnt);
|
|
|
|
#endif
|
|
|
|
c->xruns++;
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setxrun(b, sndbuf_getxrun(b) + cnt);
|
|
|
|
return sndbuf_dispose(b, NULL, cnt);
|
2000-05-26 21:55:13 +00:00
|
|
|
}
|
2006-01-01 18:16:13 +00:00
|
|
|
#endif
|
2000-05-26 21:55:13 +00:00
|
|
|
|
1999-12-13 03:29:09 +00:00
|
|
|
/*
|
2001-08-29 02:49:54 +00:00
|
|
|
* Feed new data from the read buffer. Can be called in the bottom half.
|
1999-12-13 03:29:09 +00:00
|
|
|
*/
|
1999-12-29 03:46:54 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_rdfeed(struct pcm_channel *c)
|
1999-12-13 03:29:09 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
2001-08-29 02:49:54 +00:00
|
|
|
unsigned int ret, amt;
|
2001-02-07 06:48:28 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
DEB(
|
|
|
|
if (c->flags & CHN_F_CLOSING) {
|
|
|
|
sndbuf_dump(b, "b", 0x02);
|
|
|
|
sndbuf_dump(bs, "bs", 0x02);
|
|
|
|
})
|
1999-12-13 03:29:09 +00:00
|
|
|
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#if 0
|
2001-08-29 02:49:54 +00:00
|
|
|
amt = sndbuf_getready(b);
|
2001-10-24 12:33:21 +00:00
|
|
|
if (sndbuf_getfree(bs) < amt) {
|
2001-08-29 02:49:54 +00:00
|
|
|
c->xruns++;
|
2001-10-24 12:33:21 +00:00
|
|
|
amt = sndbuf_getfree(bs);
|
|
|
|
}
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#endif
|
|
|
|
amt = sndbuf_getfree(bs);
|
2001-08-29 02:49:54 +00:00
|
|
|
ret = (amt > 0)? sndbuf_feed(b, bs, c, c->feeder, amt) : 0;
|
1999-12-13 03:29:09 +00:00
|
|
|
|
2001-09-05 14:49:44 +00:00
|
|
|
amt = sndbuf_getready(b);
|
2006-01-01 18:16:13 +00:00
|
|
|
if (amt > 0) {
|
|
|
|
c->xruns++;
|
|
|
|
sndbuf_dispose(b, NULL, amt);
|
|
|
|
}
|
2001-08-29 02:49:54 +00:00
|
|
|
|
|
|
|
chn_wakeup(c);
|
1999-12-13 03:29:09 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
return ret;
|
1999-12-13 03:29:09 +00:00
|
|
|
}
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
void
|
|
|
|
chn_rdupdate(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
int ret;
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
KASSERT(c->direction == PCMDIR_REC, ("chn_rdupdate on bad channel"));
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
if ((c->flags & CHN_F_MAPPED) || !(c->flags & CHN_F_TRIGGERED))
|
|
|
|
return;
|
|
|
|
chn_trigger(c, PCMTRIG_EMLDMARD);
|
1999-12-29 03:46:54 +00:00
|
|
|
chn_dmaupdate(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = chn_rdfeed(c);
|
2006-01-01 18:16:13 +00:00
|
|
|
DEB(if (ret)
|
|
|
|
printf("chn_rdfeed: %d\n", ret);)
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
/* read interrupt routine. Must be called with interrupts blocked. */
|
|
|
|
static void
|
|
|
|
chn_rdintr(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
int ret;
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2001-08-29 02:49:54 +00:00
|
|
|
/* tell the driver to update the primary buffer if non-dma */
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_trigger(c, PCMTRIG_EMLDMARD);
|
2001-08-29 09:17:43 +00:00
|
|
|
/* update pointers in primary buffer */
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_dmaupdate(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
/* ...and feed from primary to secondary */
|
|
|
|
ret = chn_rdfeed(c);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2002-01-23 05:41:35 +00:00
|
|
|
* user read routine - trigger if necessary, uiomove data from secondary buffer
|
2001-03-24 23:10:29 +00:00
|
|
|
* if blocking, sleep, rinse and repeat.
|
2001-02-07 20:45:40 +00:00
|
|
|
*
|
2001-03-24 23:10:29 +00:00
|
|
|
* called externally, so must handle locking
|
1999-09-01 04:08:39 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_read(struct pcm_channel *c, struct uio *buf)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
int ret, sz, count;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
2003-11-27 19:51:44 +00:00
|
|
|
void *off;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
int t, x, togo, p;
|
2001-02-07 20:45:40 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
if (!(c->flags & CHN_F_TRIGGERED))
|
2000-10-26 20:46:58 +00:00
|
|
|
chn_start(c, 0);
|
1999-12-29 03:46:54 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = 0;
|
2001-03-25 21:43:24 +00:00
|
|
|
count = hz;
|
2001-03-24 23:10:29 +00:00
|
|
|
while (!ret && (buf->uio_resid > 0) && (count > 0)) {
|
2001-10-24 12:33:21 +00:00
|
|
|
sz = MIN(buf->uio_resid, sndbuf_getready(bs));
|
1999-12-29 03:46:54 +00:00
|
|
|
|
2001-10-24 12:33:21 +00:00
|
|
|
if (sz > 0) {
|
2003-11-27 19:51:44 +00:00
|
|
|
/*
|
|
|
|
* The following assumes that the free space in
|
|
|
|
* the buffer can never be less around the
|
|
|
|
* unlock-uiomove-lock sequence.
|
|
|
|
*/
|
|
|
|
togo = sz;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
while (ret == 0 && togo > 0) {
|
2003-11-27 19:51:44 +00:00
|
|
|
p = sndbuf_getreadyptr(bs);
|
|
|
|
t = MIN(togo, sndbuf_getsize(bs) - p);
|
|
|
|
off = sndbuf_getbufofs(bs, p);
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
ret = uiomove(off, t, buf);
|
|
|
|
CHN_LOCK(c);
|
|
|
|
togo -= t;
|
|
|
|
x = sndbuf_dispose(bs, NULL, t);
|
|
|
|
}
|
|
|
|
ret = 0;
|
2001-03-24 23:10:29 +00:00
|
|
|
} else {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (c->flags & CHN_F_NBIO)
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = EWOULDBLOCK;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
else {
|
|
|
|
ret = chn_sleep(c, "pcmrd", c->timeout);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (ret == EWOULDBLOCK) {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
count -= c->timeout;
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = 0;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
} else if (ret == ERESTART || ret == EINTR) {
|
|
|
|
c->flags |= CHN_F_ABORTING;
|
|
|
|
return ret;
|
|
|
|
} else
|
2001-10-24 12:33:21 +00:00
|
|
|
count = hz;
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
2000-06-20 23:42:08 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (count <= 0) {
|
2001-03-25 21:43:24 +00:00
|
|
|
c->flags |= CHN_F_DEAD;
|
2001-09-05 16:28:41 +00:00
|
|
|
printf("%s: record interrupt timeout, channel dead\n", c->name);
|
2001-03-25 21:43:24 +00:00
|
|
|
}
|
2000-06-06 22:30:22 +00:00
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_intr(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCK(c);
|
2001-08-29 02:49:54 +00:00
|
|
|
c->interrupts++;
|
2000-04-17 17:06:47 +00:00
|
|
|
if (c->direction == PCMDIR_PLAY)
|
|
|
|
chn_wrintr(c);
|
|
|
|
else
|
|
|
|
chn_rdintr(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_UNLOCK(c);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
2000-05-26 21:55:13 +00:00
|
|
|
u_int32_t
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_start(struct pcm_channel *c, int force)
|
2000-05-26 21:55:13 +00:00
|
|
|
{
|
2001-08-29 09:17:43 +00:00
|
|
|
u_int32_t i, j;
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
/* if we're running, or if we're prevented from triggering, bail */
|
2001-08-23 11:58:38 +00:00
|
|
|
if ((c->flags & CHN_F_TRIGGERED) || ((c->flags & CHN_F_NOTRIGGER) && !force))
|
2001-03-24 23:10:29 +00:00
|
|
|
return EINVAL;
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (force) {
|
|
|
|
i = 1;
|
|
|
|
j = 0;
|
|
|
|
} else {
|
|
|
|
if (c->direction == PCMDIR_REC) {
|
|
|
|
i = sndbuf_getfree(bs);
|
|
|
|
j = sndbuf_getready(b);
|
|
|
|
} else {
|
|
|
|
struct snd_dbuf *pb;
|
|
|
|
|
|
|
|
i = sndbuf_getready(bs);
|
|
|
|
|
|
|
|
pb = BUF_PARENT(c, b);
|
|
|
|
j = min(sndbuf_xbytes(sndbuf_getsize(pb), pb, bs),
|
|
|
|
sndbuf_getsize(bs));
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (snd_verbose > 3 && SLIST_EMPTY(&c->children))
|
|
|
|
printf("%s: PCMDIR_%s (%s) threshold i=%d j=%d\n",
|
|
|
|
__func__,
|
|
|
|
(c->direction == PCMDIR_PLAY) ? "PLAY" : "REC",
|
|
|
|
(c->flags & CHN_F_VIRTUAL) ? "virtual" : "hardware",
|
|
|
|
i, j);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (i >= j) {
|
|
|
|
c->flags |= CHN_F_TRIGGERED;
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setrun(b, 1);
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
c->feedcount = (c->flags & CHN_F_CLOSING) ? 2 : 0;
|
|
|
|
c->interrupts = 0;
|
2001-08-29 09:17:43 +00:00
|
|
|
c->xruns = 0;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (c->direction == PCMDIR_PLAY && c->parentchannel == NULL) {
|
|
|
|
chn_wrfeed(c);
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: %s starting! (%s) (ready=%d "
|
|
|
|
"force=%d i=%d j=%d intrtimeout=%u)\n",
|
|
|
|
__func__,
|
|
|
|
(c->flags & CHN_F_HAS_VCHAN) ?
|
|
|
|
"VCHAN" : "HW",
|
|
|
|
(c->flags & CHN_F_CLOSING) ? "closing" :
|
|
|
|
"running",
|
|
|
|
sndbuf_getready(b),
|
|
|
|
force, i, j, c->timeout);
|
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_trigger(c, PCMTRIG_START);
|
|
|
|
return 0;
|
2000-05-26 21:55:13 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
|
|
|
|
return 0;
|
2000-05-26 21:55:13 +00:00
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
void
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_resetbuf(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2000-04-15 05:04:12 +00:00
|
|
|
c->blocks = 0;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
sndbuf_reset(b);
|
|
|
|
sndbuf_reset(bs);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
1999-12-29 03:46:54 +00:00
|
|
|
* chn_sync waits until the space in the given channel goes above
|
1999-09-01 04:08:39 +00:00
|
|
|
* a threshold. The threshold is checked against fl or rl respectively.
|
|
|
|
* Assume that the condition can become true, do not check here...
|
|
|
|
*/
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_sync(struct pcm_channel *c, int threshold)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
int ret, count, hcount, minflush, resid, residp;
|
|
|
|
struct snd_dbuf *b, *bs;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2003-08-18 23:18:47 +00:00
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (c->flags & (CHN_F_DEAD | CHN_F_ABORTING))
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (c->direction != PCMDIR_PLAY)
|
|
|
|
return EINVAL;
|
|
|
|
|
|
|
|
bs = c->bufsoft;
|
|
|
|
|
2003-08-18 23:18:47 +00:00
|
|
|
/* if we haven't yet started and nothing is buffered, else start*/
|
|
|
|
if (!(c->flags & CHN_F_TRIGGERED)) {
|
|
|
|
if (sndbuf_getready(bs) > 0) {
|
|
|
|
ret = chn_start(c, 1);
|
|
|
|
if (ret)
|
|
|
|
return ret;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
} else
|
2003-08-18 23:18:47 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
b = BUF_PARENT(c, c->bufhard);
|
|
|
|
|
|
|
|
threshold += sndbuf_getready(b);
|
|
|
|
minflush = sndbuf_xbytes(threshold, b, bs);
|
|
|
|
minflush -= minflush % sndbuf_getbps(bs);
|
|
|
|
|
|
|
|
if (minflush > 0) {
|
|
|
|
threshold = min(minflush, sndbuf_getfree(bs));
|
|
|
|
sndbuf_clear(bs, threshold);
|
|
|
|
sndbuf_acquire(bs, NULL, threshold);
|
|
|
|
minflush -= threshold;
|
|
|
|
}
|
|
|
|
|
|
|
|
resid = sndbuf_getready(bs);
|
|
|
|
residp = resid;
|
|
|
|
count = sndbuf_xbytes(minflush + resid, bs, b) / sndbuf_getblksz(b);
|
|
|
|
hcount = count;
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
while (count > 0 && resid > 0) {
|
|
|
|
ret = chn_sleep(c, "pcmsyn", c->timeout);
|
|
|
|
if (ret == ERESTART || ret == EINTR) {
|
|
|
|
c->flags |= CHN_F_ABORTING;
|
2001-03-24 23:10:29 +00:00
|
|
|
break;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
}
|
|
|
|
if (ret == 0 || ret == EWOULDBLOCK) {
|
|
|
|
resid = sndbuf_getready(bs);
|
|
|
|
if (resid == residp) {
|
|
|
|
--count;
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: [stalled] timeout=%d "
|
|
|
|
"count=%d hcount=%d "
|
|
|
|
"resid=%d minflush=%d\n",
|
|
|
|
__func__, c->timeout, count,
|
|
|
|
hcount, resid, minflush);
|
|
|
|
} else if (resid < residp && count < hcount) {
|
|
|
|
++count;
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: [resume] timeout=%d "
|
|
|
|
"count=%d hcount=%d "
|
|
|
|
"resid=%d minflush=%d\n",
|
|
|
|
__func__, c->timeout, count,
|
|
|
|
hcount, resid, minflush);
|
|
|
|
}
|
|
|
|
if (minflush > 0 && sndbuf_getfree(bs) > 0) {
|
|
|
|
threshold = min(minflush,
|
|
|
|
sndbuf_getfree(bs));
|
|
|
|
sndbuf_clear(bs, threshold);
|
|
|
|
sndbuf_acquire(bs, NULL, threshold);
|
|
|
|
resid = sndbuf_getready(bs);
|
|
|
|
minflush -= threshold;
|
|
|
|
}
|
|
|
|
residp = resid;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: timeout=%d count=%d hcount=%d resid=%d residp=%d "
|
|
|
|
"minflush=%d ret=%d\n",
|
|
|
|
__func__, c->timeout, count, hcount, resid, residp,
|
|
|
|
minflush, ret);
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
/* called externally, handle locking */
|
1999-09-01 04:08:39 +00:00
|
|
|
int
|
2001-09-12 08:38:13 +00:00
|
|
|
chn_poll(struct pcm_channel *c, int ev, struct thread *td)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
1999-12-29 03:46:54 +00:00
|
|
|
int ret;
|
|
|
|
|
2001-06-14 13:31:30 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (!(c->flags & CHN_F_MAPPED) && !(c->flags & CHN_F_TRIGGERED))
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_start(c, 1);
|
1999-12-29 03:46:54 +00:00
|
|
|
ret = 0;
|
|
|
|
if (chn_polltrigger(c) && chn_pollreset(c))
|
|
|
|
ret = ev;
|
2000-04-15 05:04:12 +00:00
|
|
|
else
|
2001-09-21 22:46:54 +00:00
|
|
|
selrecord(td, sndbuf_getsel(bs));
|
1999-12-29 03:46:54 +00:00
|
|
|
return ret;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2001-03-24 23:10:29 +00:00
|
|
|
* chn_abort terminates a running dma transfer. it may sleep up to 200ms.
|
|
|
|
* it returns the number of bytes that have not been transferred.
|
|
|
|
*
|
2001-05-27 17:22:00 +00:00
|
|
|
* called from: dsp_close, dsp_ioctl, with channel locked
|
1999-09-01 04:08:39 +00:00
|
|
|
*/
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_abort(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-06-16 21:25:10 +00:00
|
|
|
int missing = 0;
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
if (!(c->flags & CHN_F_TRIGGERED))
|
2000-06-13 23:18:43 +00:00
|
|
|
return 0;
|
1999-12-18 22:24:50 +00:00
|
|
|
c->flags |= CHN_F_ABORTING;
|
2001-03-24 23:10:29 +00:00
|
|
|
|
|
|
|
c->flags &= ~CHN_F_TRIGGERED;
|
|
|
|
/* kill the channel */
|
1999-12-18 22:24:50 +00:00
|
|
|
chn_trigger(c, PCMTRIG_ABORT);
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setrun(b, 0);
|
2001-06-07 20:06:22 +00:00
|
|
|
if (!(c->flags & CHN_F_VIRTUAL))
|
|
|
|
chn_dmaupdate(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
missing = sndbuf_getready(bs) + sndbuf_getready(b);
|
|
|
|
|
|
|
|
c->flags &= ~CHN_F_ABORTING;
|
2001-03-25 21:43:24 +00:00
|
|
|
return missing;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* this routine tries to flush the dma transfer. It is called
|
2003-08-18 23:18:47 +00:00
|
|
|
* on a close of a playback channel.
|
|
|
|
* first, if there is data in the buffer, but the dma has not yet
|
|
|
|
* begun, we need to start it.
|
|
|
|
* next, we wait for the play buffer to drain
|
|
|
|
* finally, we stop the dma.
|
2001-03-24 23:10:29 +00:00
|
|
|
*
|
2003-08-18 23:18:47 +00:00
|
|
|
* called from: dsp_close, not valid for record channels.
|
1999-09-01 04:08:39 +00:00
|
|
|
*/
|
|
|
|
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_flush(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
1999-09-01 04:08:39 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2003-08-18 23:18:47 +00:00
|
|
|
KASSERT(c->direction == PCMDIR_PLAY, ("chn_flush on bad channel"));
|
|
|
|
DEB(printf("chn_flush: c->flags 0x%08x\n", c->flags));
|
|
|
|
|
2001-03-25 21:43:24 +00:00
|
|
|
c->flags |= CHN_F_CLOSING;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
chn_sync(c, 0);
|
2001-03-24 23:10:29 +00:00
|
|
|
c->flags &= ~CHN_F_TRIGGERED;
|
|
|
|
/* kill the channel */
|
|
|
|
chn_trigger(c, PCMTRIG_ABORT);
|
|
|
|
sndbuf_setrun(b, 0);
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
c->flags &= ~CHN_F_CLOSING;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
int
|
|
|
|
fmtvalid(u_int32_t fmt, u_int32_t *fmtlist)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; fmtlist[i]; i++)
|
|
|
|
if (fmt == fmtlist[i])
|
|
|
|
return 1;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
static struct afmtstr_table default_afmtstr_table[] = {
|
|
|
|
{ "alaw", AFMT_A_LAW }, { "mulaw", AFMT_MU_LAW },
|
|
|
|
{ "u8", AFMT_U8 }, { "s8", AFMT_S8 },
|
|
|
|
{ "s16le", AFMT_S16_LE }, { "s16be", AFMT_S16_BE },
|
|
|
|
{ "u16le", AFMT_U16_LE }, { "u16be", AFMT_U16_BE },
|
|
|
|
{ "s24le", AFMT_S24_LE }, { "s24be", AFMT_S24_BE },
|
|
|
|
{ "u24le", AFMT_U24_LE }, { "u24be", AFMT_U24_BE },
|
|
|
|
{ "s32le", AFMT_S32_LE }, { "s32be", AFMT_S32_BE },
|
|
|
|
{ "u32le", AFMT_U32_LE }, { "u32be", AFMT_U32_BE },
|
|
|
|
{ NULL, 0 },
|
|
|
|
};
|
|
|
|
|
|
|
|
int
|
|
|
|
afmtstr_swap_sign(char *s)
|
|
|
|
{
|
|
|
|
if (s == NULL || strlen(s) < 2) /* full length of "s8" */
|
|
|
|
return 0;
|
|
|
|
if (*s == 's')
|
|
|
|
*s = 'u';
|
|
|
|
else if (*s == 'u')
|
|
|
|
*s = 's';
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
afmtstr_swap_endian(char *s)
|
|
|
|
{
|
|
|
|
if (s == NULL || strlen(s) < 5) /* full length of "s16le" */
|
|
|
|
return 0;
|
|
|
|
if (s[3] == 'l')
|
|
|
|
s[3] = 'b';
|
|
|
|
else if (s[3] == 'b')
|
|
|
|
s[3] = 'l';
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
u_int32_t
|
|
|
|
afmtstr2afmt(struct afmtstr_table *tbl, const char *s, int stereo)
|
|
|
|
{
|
|
|
|
size_t fsz, sz;
|
|
|
|
|
|
|
|
sz = (s == NULL) ? 0 : strlen(s);
|
|
|
|
|
|
|
|
if (sz > 1) {
|
|
|
|
|
|
|
|
if (tbl == NULL)
|
|
|
|
tbl = default_afmtstr_table;
|
|
|
|
|
|
|
|
for (; tbl->fmtstr != NULL; tbl++) {
|
|
|
|
fsz = strlen(tbl->fmtstr);
|
|
|
|
if (sz < fsz)
|
|
|
|
continue;
|
|
|
|
if (strncmp(s, tbl->fmtstr, fsz) != 0)
|
|
|
|
continue;
|
|
|
|
if (fsz == sz)
|
|
|
|
return tbl->format |
|
|
|
|
((stereo) ? AFMT_STEREO : 0);
|
|
|
|
if ((sz - fsz) < 2 || s[fsz] != ':')
|
|
|
|
break;
|
|
|
|
/*
|
|
|
|
* For now, just handle mono/stereo.
|
|
|
|
*/
|
|
|
|
if ((s[fsz + 2] == '\0' && (s[fsz + 1] == 'm' ||
|
|
|
|
s[fsz + 1] == '1')) ||
|
|
|
|
strcmp(s + fsz + 1, "mono") == 0)
|
|
|
|
return tbl->format;
|
|
|
|
if ((s[fsz + 2] == '\0' && (s[fsz + 1] == 's' ||
|
|
|
|
s[fsz + 1] == '2')) ||
|
|
|
|
strcmp(s + fsz + 1, "stereo") == 0)
|
|
|
|
return tbl->format | AFMT_STEREO;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
u_int32_t
|
|
|
|
afmt2afmtstr(struct afmtstr_table *tbl, u_int32_t afmt, char *dst,
|
|
|
|
size_t len, int type, int stereo)
|
|
|
|
{
|
|
|
|
u_int32_t fmt = 0;
|
|
|
|
char *fmtstr = NULL, *tag = "";
|
|
|
|
|
|
|
|
if (tbl == NULL)
|
|
|
|
tbl = default_afmtstr_table;
|
|
|
|
|
|
|
|
for (; tbl->format != 0; tbl++) {
|
|
|
|
if (tbl->format == 0)
|
|
|
|
break;
|
|
|
|
if ((afmt & ~AFMT_STEREO) != tbl->format)
|
|
|
|
continue;
|
|
|
|
fmt = afmt;
|
|
|
|
fmtstr = tbl->fmtstr;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (fmt != 0 && fmtstr != NULL && dst != NULL && len > 0) {
|
|
|
|
strlcpy(dst, fmtstr, len);
|
|
|
|
switch (type) {
|
|
|
|
case AFMTSTR_SIMPLE:
|
|
|
|
tag = (fmt & AFMT_STEREO) ? ":s" : ":m";
|
|
|
|
break;
|
|
|
|
case AFMTSTR_NUM:
|
|
|
|
tag = (fmt & AFMT_STEREO) ? ":2" : ":1";
|
|
|
|
break;
|
|
|
|
case AFMTSTR_FULL:
|
|
|
|
tag = (fmt & AFMT_STEREO) ? ":stereo" : ":mono";
|
|
|
|
break;
|
|
|
|
case AFMTSTR_NONE:
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (strlen(tag) > 0 && ((stereo && !(fmt & AFMT_STEREO)) || \
|
|
|
|
(!stereo && (fmt & AFMT_STEREO))))
|
|
|
|
strlcat(dst, tag, len);
|
|
|
|
}
|
|
|
|
|
|
|
|
return fmt;
|
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_reset(struct pcm_channel *c, u_int32_t fmt)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2002-01-23 05:21:27 +00:00
|
|
|
int hwspd, r;
|
2000-04-15 05:04:12 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
c->feedcount = 0;
|
1999-09-01 04:08:39 +00:00
|
|
|
c->flags &= CHN_F_RESET;
|
2001-08-29 02:49:54 +00:00
|
|
|
c->interrupts = 0;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
c->timeout = 1;
|
2001-08-29 02:49:54 +00:00
|
|
|
c->xruns = 0;
|
2002-01-23 05:21:27 +00:00
|
|
|
|
|
|
|
r = CHANNEL_RESET(c->methods, c->devinfo);
|
|
|
|
if (fmt != 0) {
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#if 0
|
2000-11-18 03:43:04 +00:00
|
|
|
hwspd = DSP_DEFAULT_SPEED;
|
2001-08-27 01:02:13 +00:00
|
|
|
/* only do this on a record channel until feederbuilder works */
|
|
|
|
if (c->direction == PCMDIR_REC)
|
|
|
|
RANGE(hwspd, chn_getcaps(c)->minspeed, chn_getcaps(c)->maxspeed);
|
2000-11-18 03:43:04 +00:00
|
|
|
c->speed = hwspd;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
#endif
|
|
|
|
hwspd = chn_getcaps(c)->minspeed;
|
|
|
|
c->speed = hwspd;
|
2000-11-18 03:43:04 +00:00
|
|
|
|
2002-01-23 05:21:27 +00:00
|
|
|
if (r == 0)
|
|
|
|
r = chn_setformat(c, fmt);
|
2000-05-26 21:55:13 +00:00
|
|
|
if (r == 0)
|
2000-11-18 03:43:04 +00:00
|
|
|
r = chn_setspeed(c, hwspd);
|
2005-10-02 15:37:40 +00:00
|
|
|
#if 0
|
2000-05-26 21:55:13 +00:00
|
|
|
if (r == 0)
|
|
|
|
r = chn_setvolume(c, 100, 100);
|
2005-10-02 15:37:40 +00:00
|
|
|
#endif
|
2000-04-17 17:06:47 +00:00
|
|
|
}
|
2002-01-23 05:21:27 +00:00
|
|
|
if (r == 0)
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
r = chn_setlatency(c, chn_latency);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (r == 0) {
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_resetbuf(c);
|
2002-01-23 05:21:27 +00:00
|
|
|
r = CHANNEL_RESETDONE(c->methods, c->devinfo);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
return r;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int
|
2004-02-28 19:47:02 +00:00
|
|
|
chn_init(struct pcm_channel *c, void *devinfo, int dir, int direction)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2000-12-18 01:36:41 +00:00
|
|
|
struct feeder_class *fc;
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b, *bs;
|
2002-01-23 05:21:27 +00:00
|
|
|
int ret;
|
1999-12-13 03:29:09 +00:00
|
|
|
|
2003-12-05 02:08:13 +00:00
|
|
|
chn_lockinit(c, dir);
|
2001-08-29 02:49:54 +00:00
|
|
|
|
2002-01-23 05:21:27 +00:00
|
|
|
b = NULL;
|
|
|
|
bs = NULL;
|
|
|
|
c->devinfo = NULL;
|
2000-12-18 01:36:41 +00:00
|
|
|
c->feeder = NULL;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
c->latency = -1;
|
|
|
|
c->timeout = 1;
|
2002-01-23 05:21:27 +00:00
|
|
|
|
2004-01-28 08:02:15 +00:00
|
|
|
ret = ENOMEM;
|
|
|
|
b = sndbuf_create(c->dev, c->name, "primary", c);
|
|
|
|
if (b == NULL)
|
|
|
|
goto out;
|
|
|
|
bs = sndbuf_create(c->dev, c->name, "secondary", c);
|
|
|
|
if (bs == NULL)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
CHN_LOCK(c);
|
|
|
|
|
2002-01-23 05:21:27 +00:00
|
|
|
ret = EINVAL;
|
2000-12-18 01:36:41 +00:00
|
|
|
fc = feeder_getclass(NULL);
|
|
|
|
if (fc == NULL)
|
2002-01-23 05:21:27 +00:00
|
|
|
goto out;
|
2000-12-18 01:36:41 +00:00
|
|
|
if (chn_addfeeder(c, fc, NULL))
|
2002-01-23 05:21:27 +00:00
|
|
|
goto out;
|
2000-08-20 22:18:56 +00:00
|
|
|
|
2004-01-28 08:02:15 +00:00
|
|
|
/*
|
|
|
|
* XXX - sndbuf_setup() & sndbuf_resize() expect to be called
|
|
|
|
* with the channel unlocked because they are also called
|
|
|
|
* from driver methods that don't know about locking
|
|
|
|
*/
|
|
|
|
CHN_UNLOCK(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setup(bs, NULL, 0);
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_LOCK(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
c->bufhard = b;
|
|
|
|
c->bufsoft = bs;
|
1999-09-01 04:08:39 +00:00
|
|
|
c->flags = 0;
|
2000-09-23 22:11:32 +00:00
|
|
|
c->feederflags = 0;
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
c->sm = NULL;
|
2002-01-23 05:21:27 +00:00
|
|
|
|
|
|
|
ret = ENODEV;
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_UNLOCK(c); /* XXX - Unlock for CHANNEL_INIT() malloc() call */
|
2004-02-28 19:47:02 +00:00
|
|
|
c->devinfo = CHANNEL_INIT(c->methods, devinfo, b, c, direction);
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_LOCK(c);
|
2002-01-23 05:21:27 +00:00
|
|
|
if (c->devinfo == NULL)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
ret = ENOMEM;
|
|
|
|
if ((sndbuf_getsize(b) == 0) && ((c->flags & CHN_F_VIRTUAL) == 0))
|
|
|
|
goto out;
|
|
|
|
|
2004-02-28 19:47:02 +00:00
|
|
|
ret = chn_setdir(c, direction);
|
2002-01-23 05:21:27 +00:00
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
ret = sndbuf_setfmt(b, AFMT_U8);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
|
|
|
ret = sndbuf_setfmt(bs, AFMT_U8);
|
|
|
|
if (ret)
|
|
|
|
goto out;
|
|
|
|
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
/**
|
|
|
|
* @todo Should this be moved somewhere else? The primary buffer
|
|
|
|
* is allocated by the driver or via DMA map setup, and tmpbuf
|
|
|
|
* seems to only come into existence in sndbuf_resize().
|
|
|
|
*/
|
|
|
|
if (c->direction == PCMDIR_PLAY) {
|
|
|
|
bs->sl = sndbuf_getmaxsize(bs);
|
|
|
|
bs->shadbuf = malloc(bs->sl, M_DEVBUF, M_NOWAIT);
|
|
|
|
if (bs->shadbuf == NULL) {
|
|
|
|
ret = ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
2002-01-23 05:21:27 +00:00
|
|
|
|
|
|
|
out:
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_UNLOCK(c);
|
2002-01-23 05:21:27 +00:00
|
|
|
if (ret) {
|
|
|
|
if (c->devinfo) {
|
|
|
|
if (CHANNEL_FREE(c->methods, c->devinfo))
|
|
|
|
sndbuf_free(b);
|
|
|
|
}
|
|
|
|
if (bs)
|
|
|
|
sndbuf_destroy(bs);
|
|
|
|
if (b)
|
|
|
|
sndbuf_destroy(b);
|
|
|
|
c->flags |= CHN_F_DEAD;
|
|
|
|
chn_lockdestroy(c);
|
|
|
|
|
|
|
|
return ret;
|
2001-03-24 23:10:29 +00:00
|
|
|
}
|
1999-12-29 03:46:54 +00:00
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2000-09-01 20:09:24 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_kill(struct pcm_channel *c)
|
2000-09-01 20:09:24 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
|
|
|
|
2000-09-09 19:21:04 +00:00
|
|
|
if (c->flags & CHN_F_TRIGGERED)
|
|
|
|
chn_trigger(c, PCMTRIG_ABORT);
|
2000-09-01 20:09:24 +00:00
|
|
|
while (chn_removefeeder(c) == 0);
|
2000-12-18 01:36:41 +00:00
|
|
|
if (CHANNEL_FREE(c->methods, c->devinfo))
|
2002-01-25 02:36:22 +00:00
|
|
|
sndbuf_free(b);
|
2000-09-01 20:09:24 +00:00
|
|
|
c->flags |= CHN_F_DEAD;
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_destroy(bs);
|
|
|
|
sndbuf_destroy(b);
|
|
|
|
chn_lockdestroy(c);
|
2000-09-01 20:09:24 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_setdir(struct pcm_channel *c, int dir)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2005-06-10 21:33:14 +00:00
|
|
|
#ifdef DEV_ISA
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
2005-06-10 21:33:14 +00:00
|
|
|
#endif
|
1999-12-29 03:46:54 +00:00
|
|
|
int r;
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
1999-09-01 04:08:39 +00:00
|
|
|
c->direction = dir;
|
2000-12-18 01:36:41 +00:00
|
|
|
r = CHANNEL_SETDIR(c->methods, c->devinfo, c->direction);
|
2005-06-10 21:33:14 +00:00
|
|
|
#ifdef DEV_ISA
|
2003-02-07 14:05:34 +00:00
|
|
|
if (!r && SND_DMA(b))
|
|
|
|
sndbuf_dmasetdir(b, c->direction);
|
2005-06-10 21:33:14 +00:00
|
|
|
#endif
|
1999-12-29 03:46:54 +00:00
|
|
|
return r;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_setvolume(struct pcm_channel *c, int left, int right)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2003-08-18 23:18:47 +00:00
|
|
|
/* should add a feeder for volume changing if channel returns -1 */
|
2005-10-02 15:37:40 +00:00
|
|
|
if (left > 100)
|
|
|
|
left = 100;
|
|
|
|
if (left < 0)
|
|
|
|
left = 0;
|
|
|
|
if (right > 100)
|
|
|
|
right = 100;
|
|
|
|
if (right < 0)
|
|
|
|
right = 0;
|
|
|
|
c->volume = left | (right << 8);
|
2001-03-25 21:43:24 +00:00
|
|
|
return 0;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
static u_int32_t
|
|
|
|
round_pow2(u_int32_t v)
|
|
|
|
{
|
|
|
|
u_int32_t ret;
|
|
|
|
|
|
|
|
if (v < 2)
|
|
|
|
v = 2;
|
|
|
|
ret = 0;
|
|
|
|
while (v >> ret)
|
|
|
|
ret++;
|
|
|
|
ret = 1 << (ret - 1);
|
|
|
|
while (ret < v)
|
|
|
|
ret <<= 1;
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* 4Front call it DSP Policy, while we call it "Latency Profile". The idea
|
|
|
|
* is to keep 2nd buffer short so that it doesn't cause long queue during
|
|
|
|
* buffer transfer.
|
|
|
|
*
|
|
|
|
* Latency reference table for 48khz stereo 16bit: (PLAY)
|
|
|
|
*
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | Latency | Blockcount | Blocksize | Buffersize |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 0 | 2 | 64 | 128 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 1 | 4 | 128 | 512 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 2 | 8 | 512 | 4096 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 3 | 16 | 512 | 8192 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 4 | 32 | 512 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 5 | 32 | 1024 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 6 | 16 | 2048 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 7 | 8 | 4096 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 8 | 4 | 8192 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 9 | 2 | 16384 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 10 | 2 | 32768 | 65536 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
*
|
|
|
|
* Recording need a different reference table. All we care is
|
|
|
|
* gobbling up everything within reasonable buffering threshold.
|
|
|
|
*
|
|
|
|
* Latency reference table for 48khz stereo 16bit: (REC)
|
|
|
|
*
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | Latency | Blockcount | Blocksize | Buffersize |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 0 | 512 | 32 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 1 | 256 | 64 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 2 | 128 | 128 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 3 | 64 | 256 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 4 | 32 | 512 | 16384 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 5 | 32 | 1024 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 6 | 16 | 2048 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 7 | 8 | 4096 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 8 | 4 | 8192 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 9 | 2 | 16384 | 32768 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
* | 10 | 2 | 32768 | 65536 |
|
|
|
|
* +---------+------------+-----------+------------+
|
|
|
|
*
|
|
|
|
* Calculations for other data rate are entirely based on these reference
|
|
|
|
* tables. For normal operation, Latency 5 seems give the best, well
|
|
|
|
* balanced performance for typical workload. Anything below 5 will
|
|
|
|
* eat up CPU to keep up with increasing context switches because of
|
|
|
|
* shorter buffer space and usually require the application to handle it
|
|
|
|
* aggresively through possibly real time programming technique.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
#define CHN_LATENCY_PBLKCNT_REF \
|
|
|
|
{{1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 1}, \
|
|
|
|
{1, 2, 3, 4, 5, 5, 4, 3, 2, 1, 1}}
|
|
|
|
#define CHN_LATENCY_PBUFSZ_REF \
|
|
|
|
{{7, 9, 12, 13, 14, 15, 15, 15, 15, 15, 16}, \
|
|
|
|
{11, 12, 13, 14, 15, 16, 16, 16, 16, 16, 17}}
|
|
|
|
|
|
|
|
#define CHN_LATENCY_RBLKCNT_REF \
|
|
|
|
{{9, 8, 7, 6, 5, 5, 4, 3, 2, 1, 1}, \
|
|
|
|
{9, 8, 7, 6, 5, 5, 4, 3, 2, 1, 1}}
|
|
|
|
#define CHN_LATENCY_RBUFSZ_REF \
|
|
|
|
{{14, 14, 14, 14, 14, 15, 15, 15, 15, 15, 16}, \
|
|
|
|
{15, 15, 15, 15, 15, 16, 16, 16, 16, 16, 17}}
|
|
|
|
|
|
|
|
#define CHN_LATENCY_DATA_REF 192000 /* 48khz stereo 16bit ~ 48000 x 2 x 2 */
|
|
|
|
|
|
|
|
static int
|
|
|
|
chn_calclatency(int dir, int latency, int bps, u_int32_t datarate,
|
|
|
|
u_int32_t max, int *rblksz, int *rblkcnt)
|
|
|
|
{
|
|
|
|
u_int32_t bufsz;
|
|
|
|
int blksz, blkcnt;
|
|
|
|
static int pblkcnts[CHN_LATENCY_PROFILE_MAX+1][CHN_LATENCY_MAX+1] =
|
|
|
|
CHN_LATENCY_PBLKCNT_REF;
|
|
|
|
static int pbufszs[CHN_LATENCY_PROFILE_MAX+1][CHN_LATENCY_MAX+1] =
|
|
|
|
CHN_LATENCY_PBUFSZ_REF;
|
|
|
|
static int rblkcnts[CHN_LATENCY_PROFILE_MAX+1][CHN_LATENCY_MAX+1] =
|
|
|
|
CHN_LATENCY_RBLKCNT_REF;
|
|
|
|
static int rbufszs[CHN_LATENCY_PROFILE_MAX+1][CHN_LATENCY_MAX+1] =
|
|
|
|
CHN_LATENCY_RBUFSZ_REF;
|
|
|
|
|
|
|
|
if (CHN_LATENCY_MIN != 0 || CHN_LATENCY_MAX != 10 ||
|
|
|
|
latency < CHN_LATENCY_MIN || latency > CHN_LATENCY_MAX ||
|
|
|
|
bps < 1 || datarate < 1 ||
|
|
|
|
!(dir == PCMDIR_PLAY || dir == PCMDIR_REC)) {
|
|
|
|
if (rblksz != NULL)
|
|
|
|
*rblksz = CHN_2NDBUFMAXSIZE >> 1;
|
|
|
|
if (rblkcnt != NULL)
|
|
|
|
*rblkcnt = 2;
|
|
|
|
printf("%s: FAILED dir=%d latency=%d bps=%d "
|
|
|
|
"datarate=%u max=%u\n",
|
|
|
|
__func__, dir, latency, bps, datarate, max);
|
|
|
|
return CHN_2NDBUFMAXSIZE;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dir == PCMDIR_PLAY) {
|
|
|
|
blkcnt = pblkcnts[chn_latency_profile][latency];
|
|
|
|
bufsz = pbufszs[chn_latency_profile][latency];
|
|
|
|
} else {
|
|
|
|
blkcnt = rblkcnts[chn_latency_profile][latency];
|
|
|
|
bufsz = rbufszs[chn_latency_profile][latency];
|
|
|
|
}
|
|
|
|
bufsz = snd_xbytes(1 << bufsz, CHN_LATENCY_DATA_REF, datarate);
|
|
|
|
if (bufsz > max)
|
|
|
|
bufsz = max;
|
|
|
|
if (bufsz < 32)
|
|
|
|
bufsz = 32;
|
|
|
|
blksz = bufsz >> blkcnt;
|
|
|
|
blksz -= blksz % bps;
|
|
|
|
while (blksz < 16 || blksz < bps)
|
|
|
|
blksz += bps;
|
|
|
|
while ((blksz << blkcnt) > bufsz && blkcnt > 1)
|
|
|
|
blkcnt--;
|
|
|
|
if (rblksz != NULL)
|
|
|
|
*rblksz = blksz;
|
|
|
|
if (rblkcnt != NULL)
|
|
|
|
*rblkcnt = 1 << blkcnt;
|
|
|
|
|
|
|
|
return blksz << blkcnt;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Note that there is no strict requirement to align blksz to the
|
|
|
|
* nearest ^2, except for hardware CHANNEL_SETBLOCKSIZE. If the application
|
|
|
|
* trying to act smarter and requesting for specific blksz/blkcnt, so be it.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
chn_resizebuf(struct pcm_channel *c, int latency,
|
|
|
|
int blkcnt, int blksz)
|
|
|
|
{
|
|
|
|
struct snd_dbuf *b, *bs, *pb;
|
|
|
|
int sblksz, sblkcnt, hblksz, limit = 1;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
|
|
|
|
if (!CANCHANGE(c) || (c->flags & CHN_F_MAPPED) ||
|
|
|
|
!(c->direction == PCMDIR_PLAY || c->direction == PCMDIR_REC))
|
|
|
|
return EINVAL;
|
|
|
|
|
|
|
|
if (latency == -1) {
|
|
|
|
c->latency = -1;
|
|
|
|
latency = chn_latency;
|
|
|
|
} else if (latency == -2) {
|
|
|
|
latency = c->latency;
|
|
|
|
if (latency < CHN_LATENCY_MIN || latency > CHN_LATENCY_MAX)
|
|
|
|
latency = chn_latency;
|
|
|
|
} else if (latency < CHN_LATENCY_MIN || latency > CHN_LATENCY_MAX)
|
|
|
|
return EINVAL;
|
|
|
|
else {
|
|
|
|
c->latency = latency;
|
|
|
|
limit = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
bs = c->bufsoft;
|
|
|
|
b = c->bufhard;
|
|
|
|
|
|
|
|
if (!(blksz == 0 || blkcnt == -1) &&
|
|
|
|
(blksz < 16 || blksz < sndbuf_getbps(bs) || blkcnt < 2 ||
|
|
|
|
(blksz * blkcnt) > CHN_2NDBUFMAXSIZE))
|
|
|
|
return EINVAL;
|
|
|
|
|
|
|
|
chn_calclatency(c->direction, latency, sndbuf_getbps(bs),
|
|
|
|
sndbuf_getbps(bs) * sndbuf_getspd(bs), CHN_2NDBUFMAXSIZE,
|
|
|
|
&sblksz, &sblkcnt);
|
|
|
|
|
|
|
|
if (blksz == 0 || blkcnt == -1) {
|
|
|
|
if (blkcnt == -1)
|
|
|
|
c->flags &= ~CHN_F_HAS_SIZE;
|
|
|
|
if (c->flags & CHN_F_HAS_SIZE) {
|
|
|
|
blksz = sndbuf_getblksz(bs);
|
|
|
|
blkcnt = sndbuf_getblkcnt(bs);
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
c->flags |= CHN_F_HAS_SIZE;
|
|
|
|
|
|
|
|
if (c->flags & CHN_F_HAS_SIZE) {
|
|
|
|
/*
|
|
|
|
* The application has requested their own blksz/blkcnt.
|
|
|
|
* Just obey with it, and let them toast alone. We can
|
|
|
|
* clamp it to the nearest latency profile, but that would
|
|
|
|
* defeat the purpose of having custom control. The least
|
|
|
|
* we can do is round it to the nearest ^2 and align it.
|
|
|
|
*/
|
|
|
|
sblksz = round_pow2(blksz);
|
|
|
|
sblksz -= sblksz % sndbuf_getbps(bs);
|
|
|
|
sblkcnt = blkcnt;
|
|
|
|
while (sblksz < 16 || sblksz < sndbuf_getbps(bs))
|
|
|
|
sblksz += sndbuf_getbps(bs);
|
|
|
|
if (snd_verbose > 3 && !(blksz == 0 || blkcnt == -1))
|
|
|
|
printf("%s: requested blksz=%d blkcnt=%d -> %d/%d\n",
|
|
|
|
__func__, blksz, blkcnt, sblksz, sblkcnt);
|
|
|
|
limit = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (c->parentchannel != NULL) {
|
|
|
|
pb = BUF_PARENT(c, NULL);
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
chn_notify(c->parentchannel, CHN_N_BLOCKSIZE);
|
|
|
|
CHN_LOCK(c);
|
|
|
|
limit = (limit != 0 && pb != NULL) ?
|
|
|
|
sndbuf_xbytes(sndbuf_getsize(pb), pb, bs) : 0;
|
|
|
|
c->timeout = c->parentchannel->timeout;
|
|
|
|
} else {
|
|
|
|
if (c->flags & CHN_F_HAS_SIZE) {
|
|
|
|
hblksz = sndbuf_xbytes(sblksz, bs, b);
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: sblksz=%d -> hblksz=%d\n",
|
|
|
|
__func__, sblksz, hblksz);
|
|
|
|
} else
|
|
|
|
chn_calclatency(c->direction, latency,
|
|
|
|
sndbuf_getbps(b),
|
|
|
|
sndbuf_getbps(b) * sndbuf_getspd(b),
|
|
|
|
CHN_2NDBUFMAXSIZE, &hblksz, NULL);
|
|
|
|
|
|
|
|
hblksz = round_pow2(hblksz);
|
|
|
|
if ((hblksz << 1) > sndbuf_getmaxsize(b))
|
|
|
|
hblksz = sndbuf_getmaxsize(b) >> 1;
|
|
|
|
hblksz -= hblksz % sndbuf_getbps(b);
|
|
|
|
while (hblksz < 16 || hblksz < sndbuf_getbps(b))
|
|
|
|
hblksz += sndbuf_getbps(b);
|
|
|
|
|
|
|
|
#if 0
|
|
|
|
hblksz = sndbuf_getmaxsize(b) >> 1;
|
|
|
|
#endif
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
sndbuf_setblksz(b, CHANNEL_SETBLOCKSIZE(c->methods,
|
|
|
|
c->devinfo, hblksz));
|
|
|
|
CHN_LOCK(c);
|
|
|
|
|
|
|
|
if (!SLIST_EMPTY(&c->children)) {
|
|
|
|
/*
|
|
|
|
* Virtual channels underneath. Set the biggest
|
|
|
|
* possible value for their mixing space.
|
|
|
|
*/
|
|
|
|
sblksz = CHN_2NDBUFMAXSIZE >> 1;
|
|
|
|
sblksz -= sblksz % sndbuf_getbps(bs);
|
|
|
|
sblkcnt = 2;
|
|
|
|
limit = 0;
|
|
|
|
} else if (limit != 0)
|
|
|
|
limit = sndbuf_xbytes(sndbuf_getsize(b), b, bs);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Interrupt timeout
|
|
|
|
*/
|
|
|
|
c->timeout = ((u_int64_t)hz * sndbuf_getblksz(b)) /
|
|
|
|
((u_int64_t)sndbuf_getspd(b) * sndbuf_getbps(b));
|
|
|
|
if (c->timeout < 1)
|
|
|
|
c->timeout = 1;
|
|
|
|
c->timeout <<= 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (limit > CHN_2NDBUFMAXSIZE)
|
|
|
|
limit = CHN_2NDBUFMAXSIZE;
|
|
|
|
|
|
|
|
hblksz = sblksz;
|
|
|
|
while ((sblksz * sblkcnt) < limit) {
|
|
|
|
sblksz += hblksz;
|
|
|
|
if ((sblksz * sblkcnt) > CHN_2NDBUFMAXSIZE) {
|
|
|
|
sblksz -= hblksz;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sndbuf_getblkcnt(bs) != sblkcnt || sndbuf_getblksz(bs) != sblksz ||
|
|
|
|
sndbuf_getsize(bs) != (sblkcnt * sblksz)) {
|
|
|
|
ret = sndbuf_remalloc(bs, sblkcnt, sblksz);
|
|
|
|
if (ret != 0) {
|
|
|
|
printf("%s: Failed: %d %d\n", __func__,
|
|
|
|
sblkcnt, sblksz);
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* OSSv4 docs: "By default OSS will set the low water level equal
|
|
|
|
* to the fragment size which is optimal in most cases."
|
|
|
|
*/
|
|
|
|
c->lw = sndbuf_getblksz(bs);
|
|
|
|
chn_resetbuf(c);
|
|
|
|
|
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("%s: PCMDIR_%s (%s) timeout=%u "
|
|
|
|
"b[%d/%d/%d] bs[%d/%d/%d] limit=%d\n",
|
|
|
|
__func__, (c->direction == PCMDIR_REC) ? "REC" : "PLAY",
|
|
|
|
(c->flags & CHN_F_VIRTUAL) ? "virtual" : "hardware",
|
|
|
|
c->timeout,
|
|
|
|
sndbuf_getsize(b), sndbuf_getblksz(b),
|
|
|
|
sndbuf_getblkcnt(b),
|
|
|
|
sndbuf_getsize(bs), sndbuf_getblksz(bs),
|
|
|
|
sndbuf_getblkcnt(bs), limit);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
chn_setlatency(struct pcm_channel *c, int latency)
|
|
|
|
{
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
/* Destroy blksz/blkcnt, enforce latency profile. */
|
|
|
|
return chn_resizebuf(c, latency, -1, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
chn_setblocksize(struct pcm_channel *c, int blkcnt, int blksz)
|
|
|
|
{
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
/* Destroy latency profile, enforce blksz/blkcnt */
|
|
|
|
return chn_resizebuf(c, -1, blkcnt, blksz);
|
|
|
|
}
|
|
|
|
|
2001-02-13 21:57:34 +00:00
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_tryspeed(struct pcm_channel *c, int speed)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct pcm_feeder *f;
|
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
2002-01-23 05:35:12 +00:00
|
|
|
struct snd_dbuf *x;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
int r, delta;
|
2000-09-23 22:11:32 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2001-05-27 17:22:00 +00:00
|
|
|
DEB(printf("setspeed, channel %s\n", c->name));
|
2000-09-23 22:11:32 +00:00
|
|
|
DEB(printf("want speed %d, ", speed));
|
2000-05-26 21:55:13 +00:00
|
|
|
if (speed <= 0)
|
|
|
|
return EINVAL;
|
1999-09-01 04:08:39 +00:00
|
|
|
if (CANCHANGE(c)) {
|
2001-03-24 23:10:29 +00:00
|
|
|
r = 0;
|
2000-09-23 22:11:32 +00:00
|
|
|
c->speed = speed;
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setspd(bs, speed);
|
|
|
|
RANGE(speed, chn_getcaps(c)->minspeed, chn_getcaps(c)->maxspeed);
|
2001-04-08 20:20:52 +00:00
|
|
|
DEB(printf("try speed %d, ", speed));
|
|
|
|
sndbuf_setspd(b, CHANNEL_SETSPEED(c->methods, c->devinfo, speed));
|
2001-05-27 17:22:00 +00:00
|
|
|
DEB(printf("got speed %d\n", sndbuf_getspd(b)));
|
2001-03-24 23:10:29 +00:00
|
|
|
|
|
|
|
delta = sndbuf_getspd(b) - sndbuf_getspd(bs);
|
2000-09-23 22:11:32 +00:00
|
|
|
if (delta < 0)
|
|
|
|
delta = -delta;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
c->feederflags &= ~(1 << FEEDER_RATE);
|
Whats New:
1. Support wide range sampling rate, as low as 1hz up to int32 max
(which is, insane) through new feeder_rate, multiple precisions
choice (32/64 bit converter). This is indeed, quite insane, but it
does give us more room and flexibility. Plenty sysctl options to
adjust resampling characteristics.
2. Support 24/32 bit pcm format conversion through new, much improved,
simplified and optimized feeder_fmt.
Changes:
1. buffer.c / dsp.c / sound.h
* Support for 24/32 AFMT.
2. feeder_rate.c
* New implementation of sampling rate conversion with 32/64 bit
precision, 1 - int32max hz (which is, ridiculous, yet very
addictive). Much improved / smarter buffer management to not
cause any missing samples at the end of conversion process
* Tunable sysctls for various aspect:
hw.snd.feeder_rate_ratemin - minimum allowable sampling rate
(default to 4000)
hw.snd.feeder_rate_ratemax - maximum allowable sampling rate
(default to 1102500)
hw.snd.feeder_rate_buffersize - conversion buffer size
(default to 8192)
hw.snd.feeder_rate_scaling - scaling / conversion method
(please refer to the source for explaination). Default to
previous implementation type.
3. feeder_fmt.c / sound.h
* New implementation, support for 24/32bit conversion, optimized,
and simplified. Few routines has been removed (8 to xlaw, 16 to
8). It just doesn't make sense.
4. channel.c
* Support for 24/32 AFMT
* Fix wrong xruns increment, causing incorrect underruns statistic
while using vchans.
5. vchan.c
* Support for 24/32 AFMT
* Proper speed / rate detection especially for fixed rate ac97.
User can override it using kernel hint:
hint.pcm.<unit>.vchanrate="xxxx".
Notes / Issues:
* Virtual Channels (vchans)
Enabling vchans can really, really help to solve overrun
issues. This is quite understandable, because it operates
entirely within its own buffering system without relying on
hardware interrupt / state. Even if you don't need vchan,
just enable single channel can help much. Few soundcards
(notably via8233x, sblive, possibly others) have their own
hardware multi channel, and this is unfortunately beyond
vchan reachability.
* The arrival of 24/32 also come with a price. Applications
that can do 24/32bit playback need to be recompiled (notably
mplayer). Use (recompiled) mplayer to experiment / test /
debug this various format using -af format=fmt. Note that
24bit seeking in mplayer is a little bit broken, sometimes
can cause silence or loud static noise. Pausing / seeking
few times can solve this problem.
You don't have to rebuild world entirely for this. Simply
copy /usr/src/sys/sys/soundcard.h to
/usr/include/sys/soundcard.h would suffice. Few drivers also
need recompilation, and this can be done via
/usr/src/sys/modules/sound/.
Support for 24bit hardware playback is beyond the scope of
this changes. That would require spessific hardware driver
changes.
* Don't expect playing 9999999999hz is a wise decision. Be
reasonable. The new feeder_rate implemention provide
flexibility, not insanity. You can easily chew up your CPU
with this kind of mind instability. Please use proper
mosquito repellent device for this obvious cracked brain
attempt. As for testing purposes, you can use (again)
mplayer to generate / play with different sampling rate. Use
something like "mplayer -af resample=192000:0:0 <files>".
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@
2005-07-31 16:16:22 +00:00
|
|
|
/*
|
|
|
|
* Used to be 500. It was too big!
|
|
|
|
*/
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (delta > feeder_rate_round)
|
2000-09-23 22:11:32 +00:00
|
|
|
c->feederflags |= 1 << FEEDER_RATE;
|
|
|
|
else
|
2001-03-24 23:10:29 +00:00
|
|
|
sndbuf_setspd(bs, sndbuf_getspd(b));
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
r = chn_buildfeeder(c);
|
|
|
|
DEB(printf("r = %d\n", r));
|
|
|
|
if (r)
|
2001-03-24 23:10:29 +00:00
|
|
|
goto out;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
if (!(c->feederflags & (1 << FEEDER_RATE)))
|
2001-03-24 23:10:29 +00:00
|
|
|
goto out;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
r = EINVAL;
|
2000-09-23 22:11:32 +00:00
|
|
|
f = chn_findfeeder(c, FEEDER_RATE);
|
|
|
|
DEB(printf("feedrate = %p\n", f));
|
|
|
|
if (f == NULL)
|
2001-03-24 23:10:29 +00:00
|
|
|
goto out;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2002-01-23 05:35:12 +00:00
|
|
|
x = (c->direction == PCMDIR_REC)? b : bs;
|
|
|
|
r = FEEDER_SET(f, FEEDRATE_SRC, sndbuf_getspd(x));
|
|
|
|
DEB(printf("feeder_set(FEEDRATE_SRC, %d) = %d\n", sndbuf_getspd(x), r));
|
2000-09-23 22:11:32 +00:00
|
|
|
if (r)
|
2001-03-24 23:10:29 +00:00
|
|
|
goto out;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2002-01-23 05:35:12 +00:00
|
|
|
x = (c->direction == PCMDIR_REC)? bs : b;
|
|
|
|
r = FEEDER_SET(f, FEEDRATE_DST, sndbuf_getspd(x));
|
|
|
|
DEB(printf("feeder_set(FEEDRATE_DST, %d) = %d\n", sndbuf_getspd(x), r));
|
2001-03-24 23:10:29 +00:00
|
|
|
out:
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
if (!r)
|
|
|
|
r = CHANNEL_SETFORMAT(c->methods, c->devinfo,
|
|
|
|
sndbuf_getfmt(b));
|
|
|
|
if (!r)
|
|
|
|
sndbuf_setfmt(bs, c->format);
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (!r)
|
|
|
|
r = chn_resizebuf(c, -2, 0, 0);
|
2001-05-27 17:22:00 +00:00
|
|
|
DEB(printf("setspeed done, r = %d\n", r));
|
2001-03-25 21:43:24 +00:00
|
|
|
return r;
|
2001-03-24 23:10:29 +00:00
|
|
|
} else
|
|
|
|
return EINVAL;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_setspeed(struct pcm_channel *c, int speed)
|
2001-02-13 21:57:34 +00:00
|
|
|
{
|
|
|
|
int r, oldspeed = c->speed;
|
|
|
|
|
|
|
|
r = chn_tryspeed(c, speed);
|
|
|
|
if (r) {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("Failed to set speed %d falling back to %d\n",
|
|
|
|
speed, oldspeed);
|
2003-02-26 14:38:19 +00:00
|
|
|
r = chn_tryspeed(c, oldspeed);
|
2001-02-13 21:57:34 +00:00
|
|
|
}
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_tryformat(struct pcm_channel *c, u_int32_t fmt)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
|
|
|
struct snd_dbuf *bs = c->bufsoft;
|
2000-09-23 22:11:32 +00:00
|
|
|
int r;
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
1999-09-01 04:08:39 +00:00
|
|
|
if (CANCHANGE(c)) {
|
2000-09-23 22:11:32 +00:00
|
|
|
DEB(printf("want format %d\n", fmt));
|
2000-05-26 21:55:13 +00:00
|
|
|
c->format = fmt;
|
2000-09-23 22:11:32 +00:00
|
|
|
r = chn_buildfeeder(c);
|
2001-03-24 23:10:29 +00:00
|
|
|
if (r == 0) {
|
2002-01-23 05:21:27 +00:00
|
|
|
sndbuf_setfmt(bs, c->format);
|
2001-03-25 21:43:24 +00:00
|
|
|
chn_resetbuf(c);
|
2002-01-23 05:21:27 +00:00
|
|
|
r = CHANNEL_SETFORMAT(c->methods, c->devinfo, sndbuf_getfmt(b));
|
|
|
|
if (r == 0)
|
|
|
|
r = chn_tryspeed(c, c->speed);
|
2001-03-25 21:43:24 +00:00
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
return r;
|
|
|
|
} else
|
|
|
|
return EINVAL;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
2001-02-13 21:57:34 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_setformat(struct pcm_channel *c, u_int32_t fmt)
|
2001-02-13 21:57:34 +00:00
|
|
|
{
|
|
|
|
u_int32_t oldfmt = c->format;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
r = chn_tryformat(c, fmt);
|
|
|
|
if (r) {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (snd_verbose > 3)
|
|
|
|
printf("Format change 0x%08x failed, reverting to 0x%08x\n",
|
|
|
|
fmt, oldfmt);
|
2001-02-13 21:57:34 +00:00
|
|
|
chn_tryformat(c, oldfmt);
|
|
|
|
}
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
1999-09-01 04:08:39 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_trigger(struct pcm_channel *c, int go)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2005-06-10 21:33:14 +00:00
|
|
|
#ifdef DEV_ISA
|
2001-03-24 23:10:29 +00:00
|
|
|
struct snd_dbuf *b = c->bufhard;
|
2005-06-10 21:33:14 +00:00
|
|
|
#endif
|
2001-03-24 23:10:29 +00:00
|
|
|
int ret;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
2005-06-10 21:33:14 +00:00
|
|
|
#ifdef DEV_ISA
|
2003-02-07 14:05:34 +00:00
|
|
|
if (SND_DMA(b) && (go == PCMTRIG_EMLDMAWR || go == PCMTRIG_EMLDMARD))
|
|
|
|
sndbuf_dmabounce(b);
|
2005-06-10 21:33:14 +00:00
|
|
|
#endif
|
2001-03-24 23:10:29 +00:00
|
|
|
ret = CHANNEL_TRIGGER(c->methods, c->devinfo, go);
|
|
|
|
|
|
|
|
return ret;
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
/**
|
|
|
|
* @brief Queries sound driver for sample-aligned hardware buffer pointer index
|
|
|
|
*
|
|
|
|
* This function obtains the hardware pointer location, then aligns it to
|
|
|
|
* the current bytes-per-sample value before returning. (E.g., a channel
|
|
|
|
* running in 16 bit stereo mode would require 4 bytes per sample, so a
|
|
|
|
* hwptr value ranging from 32-35 would be returned as 32.)
|
|
|
|
*
|
|
|
|
* @param c PCM channel context
|
|
|
|
* @returns sample-aligned hardware buffer pointer index
|
|
|
|
*/
|
1999-09-01 04:08:39 +00:00
|
|
|
int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_getptr(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2006-03-22 00:34:17 +00:00
|
|
|
#if 0
|
1999-12-05 19:09:13 +00:00
|
|
|
int hwptr;
|
|
|
|
int a = (1 << c->align) - 1;
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
hwptr = (c->flags & CHN_F_TRIGGERED)? CHANNEL_GETPTR(c->methods, c->devinfo) : 0;
|
1999-12-05 19:09:13 +00:00
|
|
|
/* don't allow unaligned values in the hwa ptr */
|
2001-03-24 23:10:29 +00:00
|
|
|
#if 1
|
1999-12-05 19:09:13 +00:00
|
|
|
hwptr &= ~a ; /* Apply channel align mask */
|
2001-03-24 23:10:29 +00:00
|
|
|
#endif
|
1999-12-05 19:09:13 +00:00
|
|
|
hwptr &= DMA_ALIGN_MASK; /* Apply DMA align mask */
|
|
|
|
return hwptr;
|
2006-03-22 00:34:17 +00:00
|
|
|
#endif
|
|
|
|
int hwptr;
|
|
|
|
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
hwptr = (c->flags & CHN_F_TRIGGERED)? CHANNEL_GETPTR(c->methods, c->devinfo) : 0;
|
|
|
|
return (hwptr - (hwptr % sndbuf_getbps(c->bufhard)));
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
struct pcmchan_caps *
|
|
|
|
chn_getcaps(struct pcm_channel *c)
|
1999-09-01 04:08:39 +00:00
|
|
|
{
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2000-12-18 01:36:41 +00:00
|
|
|
return CHANNEL_GETCAPS(c->methods, c->devinfo);
|
1999-09-01 04:08:39 +00:00
|
|
|
}
|
2000-08-20 22:18:56 +00:00
|
|
|
|
|
|
|
u_int32_t
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_getformats(struct pcm_channel *c)
|
2000-08-20 22:18:56 +00:00
|
|
|
{
|
|
|
|
u_int32_t *fmtlist, fmts;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
fmtlist = chn_getcaps(c)->fmtlist;
|
|
|
|
fmts = 0;
|
|
|
|
for (i = 0; fmtlist[i]; i++)
|
|
|
|
fmts |= fmtlist[i];
|
|
|
|
|
2002-02-24 00:49:43 +00:00
|
|
|
/* report software-supported formats */
|
|
|
|
if (report_soft_formats)
|
Whats New:
1. Support wide range sampling rate, as low as 1hz up to int32 max
(which is, insane) through new feeder_rate, multiple precisions
choice (32/64 bit converter). This is indeed, quite insane, but it
does give us more room and flexibility. Plenty sysctl options to
adjust resampling characteristics.
2. Support 24/32 bit pcm format conversion through new, much improved,
simplified and optimized feeder_fmt.
Changes:
1. buffer.c / dsp.c / sound.h
* Support for 24/32 AFMT.
2. feeder_rate.c
* New implementation of sampling rate conversion with 32/64 bit
precision, 1 - int32max hz (which is, ridiculous, yet very
addictive). Much improved / smarter buffer management to not
cause any missing samples at the end of conversion process
* Tunable sysctls for various aspect:
hw.snd.feeder_rate_ratemin - minimum allowable sampling rate
(default to 4000)
hw.snd.feeder_rate_ratemax - maximum allowable sampling rate
(default to 1102500)
hw.snd.feeder_rate_buffersize - conversion buffer size
(default to 8192)
hw.snd.feeder_rate_scaling - scaling / conversion method
(please refer to the source for explaination). Default to
previous implementation type.
3. feeder_fmt.c / sound.h
* New implementation, support for 24/32bit conversion, optimized,
and simplified. Few routines has been removed (8 to xlaw, 16 to
8). It just doesn't make sense.
4. channel.c
* Support for 24/32 AFMT
* Fix wrong xruns increment, causing incorrect underruns statistic
while using vchans.
5. vchan.c
* Support for 24/32 AFMT
* Proper speed / rate detection especially for fixed rate ac97.
User can override it using kernel hint:
hint.pcm.<unit>.vchanrate="xxxx".
Notes / Issues:
* Virtual Channels (vchans)
Enabling vchans can really, really help to solve overrun
issues. This is quite understandable, because it operates
entirely within its own buffering system without relying on
hardware interrupt / state. Even if you don't need vchan,
just enable single channel can help much. Few soundcards
(notably via8233x, sblive, possibly others) have their own
hardware multi channel, and this is unfortunately beyond
vchan reachability.
* The arrival of 24/32 also come with a price. Applications
that can do 24/32bit playback need to be recompiled (notably
mplayer). Use (recompiled) mplayer to experiment / test /
debug this various format using -af format=fmt. Note that
24bit seeking in mplayer is a little bit broken, sometimes
can cause silence or loud static noise. Pausing / seeking
few times can solve this problem.
You don't have to rebuild world entirely for this. Simply
copy /usr/src/sys/sys/soundcard.h to
/usr/include/sys/soundcard.h would suffice. Few drivers also
need recompilation, and this can be done via
/usr/src/sys/modules/sound/.
Support for 24bit hardware playback is beyond the scope of
this changes. That would require spessific hardware driver
changes.
* Don't expect playing 9999999999hz is a wise decision. Be
reasonable. The new feeder_rate implemention provide
flexibility, not insanity. You can easily chew up your CPU
with this kind of mind instability. Please use proper
mosquito repellent device for this obvious cracked brain
attempt. As for testing purposes, you can use (again)
mplayer to generate / play with different sampling rate. Use
something like "mplayer -af resample=192000:0:0 <files>".
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
Tested by: multimedia@
2005-07-31 16:16:22 +00:00
|
|
|
fmts |= AFMT_MU_LAW|AFMT_A_LAW|AFMT_U32_LE|AFMT_U32_BE|
|
|
|
|
AFMT_S32_LE|AFMT_S32_BE|AFMT_U24_LE|AFMT_U24_BE|
|
|
|
|
AFMT_S24_LE|AFMT_S24_BE|AFMT_U16_LE|AFMT_U16_BE|
|
2002-02-24 00:49:43 +00:00
|
|
|
AFMT_S16_LE|AFMT_S16_BE|AFMT_U8|AFMT_S8;
|
|
|
|
|
2000-08-20 22:18:56 +00:00
|
|
|
return fmts;
|
|
|
|
}
|
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
static int
|
2001-03-24 23:10:29 +00:00
|
|
|
chn_buildfeeder(struct pcm_channel *c)
|
2000-09-23 22:11:32 +00:00
|
|
|
{
|
2000-12-18 01:36:41 +00:00
|
|
|
struct feeder_class *fc;
|
2000-09-23 22:11:32 +00:00
|
|
|
struct pcm_feederdesc desc;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
u_int32_t tmp[2], type, flags, hwfmt, *fmtlist;
|
2002-01-26 22:13:24 +00:00
|
|
|
int err;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
char fmtstr[AFMTSTR_MAXSZ];
|
2000-08-20 22:18:56 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
CHN_LOCKASSERT(c);
|
2000-09-23 22:11:32 +00:00
|
|
|
while (chn_removefeeder(c) == 0);
|
2000-12-18 01:36:41 +00:00
|
|
|
KASSERT((c->feeder == NULL), ("feeder chain not empty"));
|
2001-04-08 20:20:52 +00:00
|
|
|
|
2001-03-24 23:10:29 +00:00
|
|
|
c->align = sndbuf_getalign(c->bufsoft);
|
2001-04-08 20:20:52 +00:00
|
|
|
|
2001-05-27 17:22:00 +00:00
|
|
|
if (SLIST_EMPTY(&c->children)) {
|
|
|
|
fc = feeder_getclass(NULL);
|
2002-01-26 22:13:24 +00:00
|
|
|
KASSERT(fc != NULL, ("can't find root feeder"));
|
|
|
|
|
|
|
|
err = chn_addfeeder(c, fc, NULL);
|
|
|
|
if (err) {
|
|
|
|
DEB(printf("can't add root feeder, err %d\n", err));
|
|
|
|
|
|
|
|
return err;
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
|
|
|
c->feeder->desc->out = c->format;
|
|
|
|
} else {
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
if (c->flags & CHN_F_HAS_VCHAN) {
|
|
|
|
desc.type = FEEDER_MIXER;
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
desc.in = c->format;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
} else {
|
|
|
|
DEB(printf("can't decide which feeder type to use!\n"));
|
|
|
|
return EOPNOTSUPP;
|
|
|
|
}
|
2001-05-27 17:22:00 +00:00
|
|
|
desc.out = c->format;
|
|
|
|
desc.flags = 0;
|
|
|
|
fc = feeder_getclass(&desc);
|
|
|
|
if (fc == NULL) {
|
|
|
|
DEB(printf("can't find vchan feeder\n"));
|
2002-01-26 22:13:24 +00:00
|
|
|
|
|
|
|
return EOPNOTSUPP;
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
2002-01-26 22:13:24 +00:00
|
|
|
|
|
|
|
err = chn_addfeeder(c, fc, &desc);
|
|
|
|
if (err) {
|
|
|
|
DEB(printf("can't add vchan feeder, err %d\n", err));
|
|
|
|
|
|
|
|
return err;
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
2001-04-08 20:20:52 +00:00
|
|
|
}
|
2005-10-02 15:37:40 +00:00
|
|
|
c->feederflags &= ~(1 << FEEDER_VOLUME);
|
|
|
|
if (c->direction == PCMDIR_PLAY &&
|
|
|
|
!(c->flags & CHN_F_VIRTUAL) &&
|
2006-09-28 17:29:00 +00:00
|
|
|
c->parentsnddev && (c->parentsnddev->flags & SD_F_SOFTPCMVOL) &&
|
2005-10-02 15:37:40 +00:00
|
|
|
c->parentsnddev->mixer_dev)
|
|
|
|
c->feederflags |= 1 << FEEDER_VOLUME;
|
2000-09-23 22:11:32 +00:00
|
|
|
flags = c->feederflags;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
fmtlist = chn_getcaps(c)->fmtlist;
|
2001-04-08 20:20:52 +00:00
|
|
|
|
2003-08-18 23:18:47 +00:00
|
|
|
DEB(printf("feederflags %x\n", flags));
|
2001-04-08 20:20:52 +00:00
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
for (type = FEEDER_RATE; type <= FEEDER_LAST; type++) {
|
|
|
|
if (flags & (1 << type)) {
|
|
|
|
desc.type = type;
|
|
|
|
desc.in = 0;
|
|
|
|
desc.out = 0;
|
|
|
|
desc.flags = 0;
|
|
|
|
DEB(printf("find feeder type %d, ", type));
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
if (type == FEEDER_VOLUME || type == FEEDER_RATE) {
|
|
|
|
if (c->feeder->desc->out & AFMT_32BIT)
|
|
|
|
strlcpy(fmtstr,"s32le", sizeof(fmtstr));
|
|
|
|
else if (c->feeder->desc->out & AFMT_24BIT)
|
|
|
|
strlcpy(fmtstr, "s24le", sizeof(fmtstr));
|
|
|
|
else {
|
|
|
|
/*
|
|
|
|
* 8bit doesn't provide enough headroom
|
|
|
|
* for proper processing without
|
|
|
|
* creating too much noises. Force to
|
|
|
|
* 16bit instead.
|
|
|
|
*/
|
|
|
|
strlcpy(fmtstr, "s16le", sizeof(fmtstr));
|
|
|
|
}
|
|
|
|
if (!(c->feeder->desc->out & AFMT_8BIT) &&
|
|
|
|
c->feeder->desc->out & AFMT_BIGENDIAN)
|
|
|
|
afmtstr_swap_endian(fmtstr);
|
|
|
|
if (!(c->feeder->desc->out & (AFMT_A_LAW | AFMT_MU_LAW)) &&
|
|
|
|
!(c->feeder->desc->out & AFMT_SIGNED))
|
|
|
|
afmtstr_swap_sign(fmtstr);
|
|
|
|
desc.in = afmtstr2afmt(NULL, fmtstr, AFMTSTR_MONO_RETURN);
|
|
|
|
if (desc.in == 0)
|
|
|
|
desc.in = AFMT_S16_LE;
|
|
|
|
/* feeder_volume need stereo processing */
|
|
|
|
if (type == FEEDER_VOLUME ||
|
|
|
|
c->feeder->desc->out & AFMT_STEREO)
|
|
|
|
desc.in |= AFMT_STEREO;
|
|
|
|
desc.out = desc.in;
|
|
|
|
fc = feeder_getclass(&desc);
|
|
|
|
if (fc != NULL && fc->desc != NULL)
|
|
|
|
desc.flags = fc->desc->flags;
|
|
|
|
} else {
|
|
|
|
fc = feeder_getclass(&desc);
|
|
|
|
if (fc != NULL && fc->desc != NULL)
|
|
|
|
desc = *fc->desc;
|
|
|
|
}
|
update code dealing with snd_dbuf objects to do so using a functional interface
modify chn_setblocksize() to pick a default soft-blocksize appropriate to the
sample rate and format in use. it will aim for a power of two size small
enough to generate block sizes of at most 20ms. it will also set the
hard-blocksize taking into account rate/format conversions in use.
update drivers to implement setblocksize correctly:
updated, tested: sb16, emu10k1, maestro, solo
updated, untested: ad1816, ess, mss, sb8, csa
not updated: ds1, es137x, fm801, neomagic, t4dwave, via82c686
i lack hardware to test: ad1816, csa, fm801, neomagic
others will be updated/tested in the next few days.
2000-12-23 03:16:13 +00:00
|
|
|
DEB(printf("got %p\n", fc));
|
2001-04-08 20:20:52 +00:00
|
|
|
if (fc == NULL) {
|
|
|
|
DEB(printf("can't find required feeder type %d\n", type));
|
2002-01-26 22:13:24 +00:00
|
|
|
|
|
|
|
return EOPNOTSUPP;
|
2001-04-08 20:20:52 +00:00
|
|
|
}
|
|
|
|
|
2006-01-22 15:06:49 +00:00
|
|
|
DEB(printf("build fmtchain from 0x%08x to 0x%08x: ", c->feeder->desc->out, fc->desc->in));
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
tmp[0] = desc.in;
|
2006-01-22 15:06:49 +00:00
|
|
|
tmp[1] = 0;
|
|
|
|
if (chn_fmtchain(c, tmp) == 0) {
|
|
|
|
DEB(printf("failed\n"));
|
|
|
|
|
|
|
|
return ENODEV;
|
2000-09-23 22:11:32 +00:00
|
|
|
}
|
2006-01-22 15:06:49 +00:00
|
|
|
DEB(printf("ok\n"));
|
2001-04-08 20:20:52 +00:00
|
|
|
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
err = chn_addfeeder(c, fc, &desc);
|
2002-01-26 22:13:24 +00:00
|
|
|
if (err) {
|
2003-08-18 23:18:47 +00:00
|
|
|
DEB(printf("can't add feeder %p, output 0x%x, err %d\n", fc, fc->desc->out, err));
|
2002-01-26 22:13:24 +00:00
|
|
|
|
|
|
|
return err;
|
2001-04-08 20:20:52 +00:00
|
|
|
}
|
2003-08-18 23:18:47 +00:00
|
|
|
DEB(printf("added feeder %p, output 0x%x\n", fc, c->feeder->desc->out));
|
2000-09-23 22:11:32 +00:00
|
|
|
}
|
|
|
|
}
|
2001-04-08 20:20:52 +00:00
|
|
|
|
2006-01-22 15:06:49 +00:00
|
|
|
if (c->direction == PCMDIR_REC) {
|
|
|
|
tmp[0] = c->format;
|
|
|
|
tmp[1] = 0;
|
|
|
|
hwfmt = chn_fmtchain(c, tmp);
|
|
|
|
} else
|
|
|
|
hwfmt = chn_fmtchain(c, fmtlist);
|
2001-04-08 20:20:52 +00:00
|
|
|
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
if (hwfmt == 0 || !fmtvalid(hwfmt, fmtlist)) {
|
2006-01-22 15:06:49 +00:00
|
|
|
DEB(printf("Invalid hardware format: 0x%08x\n", hwfmt));
|
2002-01-26 22:13:24 +00:00
|
|
|
return ENODEV;
|
- channel.h
* New definition CHN_F_HAS_VCHAN.
- channel.c
* Use CHN_F_HAS_VCHAN to mark channel with vchan capability instead
of relying on SLIST_EMPTY(&channel->children) == true for better
clarification and future possible usages of children (like
'slave' channel).
* Various fixes, including blocksize / format bps allignment,
better 24bit seeking (mplayer, others).
* Improve format chain building, it's now possible to record something
to a format non-native to the soundcard through various feeder format
converters or to higher sampling rate. This also gains another feature,
like doing vchan mixing on non s16le soundcard such as sb8.
- sound.c
* Increase robustness within various function that handle vchan
creation / termination (these function need a total rewrite, but
that would cause other major rewrite within various places too!).
As far as its robustness can be guaranteed, leave it as is.
* Optimize channel ordering, prefer *real* hardware playback
channels over virtual channels. cat /dev/sndstat should look
better.
* Increase sndstat verbosity to include bufsoft/bufhard allocation.
- vchan.c
* Fix LOR 119.
- http://sources.zabbadoz.net/freebsd/lor.html#119
* Reorder / increase robustness of vchan_create() / destroy().
Enforce destroy_dev() during destroy operation, fix possible
panic / dangling character device.
- http://lists.freebsd.org/pipermail/freebsd-current/2005-May/050308.html
* Tolerate a little bit more during mixing process, this should help
non s16le soundcards.
Note: Recoring in a non-native rate/format may result in overruns. A friendly
application is wavrec from audio/wavplay. The problem is under
investigation.
Submitted by: Ariff Abdullah <skywizard@MyBSD.org.my>
2005-09-10 18:10:31 +00:00
|
|
|
}
|
2002-01-23 05:10:56 +00:00
|
|
|
|
|
|
|
sndbuf_setfmt(c->bufhard, hwfmt);
|
2002-01-26 22:13:24 +00:00
|
|
|
|
2005-10-02 15:37:40 +00:00
|
|
|
if ((flags & (1 << FEEDER_VOLUME))) {
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
u_int32_t parent = SOUND_MIXER_NONE;
|
2006-09-28 17:29:00 +00:00
|
|
|
int vol, left, right;
|
|
|
|
|
|
|
|
vol = 100 | (100 << 8);
|
2005-10-02 15:37:40 +00:00
|
|
|
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
/*
|
|
|
|
* XXX This is ugly! The way mixer subs being so secretive
|
|
|
|
* about its own internals force us to use this silly
|
|
|
|
* monkey trick.
|
|
|
|
*/
|
|
|
|
if (mixer_ioctl(c->parentsnddev->mixer_dev,
|
|
|
|
MIXER_READ(SOUND_MIXER_PCM), (caddr_t)&vol, -1, NULL) != 0)
|
2006-09-28 17:29:00 +00:00
|
|
|
device_printf(c->dev, "Soft PCM Volume: Failed to read default value\n");
|
|
|
|
left = vol & 0x7f;
|
|
|
|
right = (vol >> 8) & 0x7f;
|
|
|
|
if (c->parentsnddev != NULL &&
|
|
|
|
c->parentsnddev->mixer_dev != NULL &&
|
|
|
|
c->parentsnddev->mixer_dev->si_drv1 != NULL)
|
|
|
|
parent = mix_getparent(
|
|
|
|
c->parentsnddev->mixer_dev->si_drv1,
|
|
|
|
SOUND_MIXER_PCM);
|
|
|
|
if (parent != SOUND_MIXER_NONE) {
|
|
|
|
vol = 100 | (100 << 8);
|
|
|
|
if (mixer_ioctl(c->parentsnddev->mixer_dev,
|
|
|
|
MIXER_READ(parent),
|
|
|
|
(caddr_t)&vol, -1, NULL) != 0)
|
|
|
|
device_printf(c->dev, "Soft Volume: Failed to read parent default value\n");
|
|
|
|
left = (left * (vol & 0x7f)) / 100;
|
|
|
|
right = (right * ((vol >> 8) & 0x7f)) / 100;
|
|
|
|
}
|
2005-10-02 15:37:40 +00:00
|
|
|
CHN_LOCK(c);
|
2006-09-28 17:29:00 +00:00
|
|
|
chn_setvolume(c, left, right);
|
2005-10-02 15:37:40 +00:00
|
|
|
}
|
|
|
|
|
2000-09-23 22:11:32 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2001-03-24 23:10:29 +00:00
|
|
|
|
2001-05-27 17:22:00 +00:00
|
|
|
int
|
|
|
|
chn_notify(struct pcm_channel *c, u_int32_t flags)
|
|
|
|
{
|
|
|
|
struct pcmchan_children *pce;
|
|
|
|
struct pcm_channel *child;
|
2001-06-07 20:06:22 +00:00
|
|
|
int run;
|
2001-05-27 17:22:00 +00:00
|
|
|
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_LOCK(c);
|
|
|
|
|
|
|
|
if (SLIST_EMPTY(&c->children)) {
|
|
|
|
CHN_UNLOCK(c);
|
2001-05-27 17:22:00 +00:00
|
|
|
return ENODEV;
|
2004-01-28 08:02:15 +00:00
|
|
|
}
|
2001-05-27 17:22:00 +00:00
|
|
|
|
2001-06-07 20:06:22 +00:00
|
|
|
run = (c->flags & CHN_F_TRIGGERED)? 1 : 0;
|
|
|
|
/*
|
|
|
|
* if the hwchan is running, we can't change its rate, format or
|
|
|
|
* blocksize
|
|
|
|
*/
|
|
|
|
if (run)
|
|
|
|
flags &= CHN_N_VOLUME | CHN_N_TRIGGER;
|
|
|
|
|
2001-05-27 17:22:00 +00:00
|
|
|
if (flags & CHN_N_RATE) {
|
|
|
|
/*
|
|
|
|
* we could do something here, like scan children and decide on
|
|
|
|
* the most appropriate rate to mix at, but we don't for now
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
if (flags & CHN_N_FORMAT) {
|
|
|
|
/*
|
|
|
|
* we could do something here, like scan children and decide on
|
|
|
|
* the most appropriate mixer feeder to use, but we don't for now
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
if (flags & CHN_N_VOLUME) {
|
|
|
|
/*
|
|
|
|
* we could do something here but we don't for now
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
if (flags & CHN_N_BLOCKSIZE) {
|
|
|
|
/*
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
* Set to default latency profile
|
2001-05-27 17:22:00 +00:00
|
|
|
*/
|
Welcome to Once-a-year Sound Mega-Commit. Enjoy numerous updates and fixes
in every sense.
General
-------
- Multichannel safe, endian safe, format safe
* Large part of critical pcm filters such as vchan.c, feeder_rate.c,
feeder_volume.c, feeder_fmt.c and feeder.c has been rewritten so that
using them does not cause the pcm data to be converted to 16bit little
endian.
* Macrosses for accessing pcm data safely are defined within sound.h in
the form of PCM_READ_* / PCM_WRITE_*
* Currently, most of them are probably limited for mono/stereo handling,
but the future addition of true multichannel will be much easier.
- Low latency operation
* Well, this require lot more works to do not just within sound driver,
but we're heading towards right direction. Buffer/block sizing within
channel.c is rewritten to calculate precise allocation for various
combination of sample/data/rate size. As a result, applying correct
SNDCTL_DSP_POLICY value will achive expected latency behaviour simmilar
to what commercial 4front driver do.
* Signal handling fix. ctrl+c of "cat /dev/zero > /dev/dsp" does not
result long delay.
* Eliminate sound truncation if the sound data is too small.
DIY:
1) Download / extract
http://people.freebsd.org/~ariff/lowlatency/shortfiles.tar.gz
2) Do a comparison between "cat state*.au > /dev/dsp" and
"for x in state*.au ; do cat $x > /dev/dsp ; done"
- there should be no "perceivable" differences.
Double close for PR kern/31445.
CAVEAT: Low latency come with (unbearable) price especially for poorly
written applications. Applications that trying to act smarter
by requesting (wrong) blocksize/blockcount will suffer the most.
Fixup samples/patches can be found at:
http://people.freebsd.org/~ariff/ports/
- Switch minimum/maximum sampling rate limit to "1" and "2016000" (48k * 42)
due to closer compatibility with 4front driver.
Discussed with: marcus@ (long time ago?)
- All driver specific sysctls in the form of "hw.snd.pcm%d.*" have been
moved to their own dev sysctl nodes, notably:
hw.snd.pcm%d.vchans -> dev.pcm.%d.vchans
Bump __FreeBSD_version.
Driver specific
---------------
- Ditto for sysctls.
- snd_atiixp, snd_es137x, snd_via8233, snd_hda
* Numerous cleanups and fixes.
* _EXPERIMENTAL_ polling mode support using simple callout_* mechanisme.
This was intended for pure debugging and latency measurement, but proven
good enough in few unexpected and rare cases (such as problematic shared
IRQ with GIANT devices - USB). Polling can be enabled/disabled through
dev.pcm.0.polling. Disabled by default.
- snd_ich
* Fix possible overflow during speed calibration. Delay final
initialization (pcm_setstatus) after calibration finished.
PR: kern/100169
Tested by: Kevin Overman <oberman@es.net>
* Inverted EAPD for few Nec VersaPro.
PR: kern/104715
Submitted by: KAWATA Masahiko <kawata@mta.biglobe.ne.jp>
Thanks to various people, notably Joel Dahl, Yuriy Tsibizov, Kevin Oberman,
those at #freebsd-azalia @ freenode and others for testing.
Joel Dahl will do the manpage update.
2006-11-26 12:24:06 +00:00
|
|
|
chn_setlatency(c, chn_latency);
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
|
|
|
if (flags & CHN_N_TRIGGER) {
|
2001-06-07 20:06:22 +00:00
|
|
|
int nrun;
|
2001-05-27 17:22:00 +00:00
|
|
|
/*
|
|
|
|
* scan the children, and figure out if any are running
|
|
|
|
* if so, we need to be running, otherwise we need to be stopped
|
|
|
|
* if we aren't in our target sstate, move to it
|
|
|
|
*/
|
2001-06-07 20:06:22 +00:00
|
|
|
nrun = 0;
|
2001-05-27 17:22:00 +00:00
|
|
|
SLIST_FOREACH(pce, &c->children, link) {
|
|
|
|
child = pce->channel;
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_LOCK(child);
|
2001-05-27 17:22:00 +00:00
|
|
|
if (child->flags & CHN_F_TRIGGERED)
|
2001-06-07 20:06:22 +00:00
|
|
|
nrun = 1;
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_UNLOCK(child);
|
2001-05-27 17:22:00 +00:00
|
|
|
}
|
2001-06-07 20:06:22 +00:00
|
|
|
if (nrun && !run)
|
2001-05-27 17:22:00 +00:00
|
|
|
chn_start(c, 1);
|
2001-06-07 20:06:22 +00:00
|
|
|
if (!nrun && run)
|
2001-05-27 17:22:00 +00:00
|
|
|
chn_abort(c);
|
|
|
|
}
|
2004-01-28 08:02:15 +00:00
|
|
|
CHN_UNLOCK(c);
|
2001-05-27 17:22:00 +00:00
|
|
|
return 0;
|
|
|
|
}
|
2004-01-28 08:02:15 +00:00
|
|
|
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
/**
|
|
|
|
* @brief Fetch array of supported discrete sample rates
|
|
|
|
*
|
|
|
|
* Wrapper for CHANNEL_GETRATES. Please see channel_if.m:getrates() for
|
|
|
|
* detailed information.
|
|
|
|
*
|
|
|
|
* @note If the operation isn't supported, this function will just return 0
|
|
|
|
* (no rates in the array), and *rates will be set to NULL. Callers
|
|
|
|
* should examine rates @b only if this function returns non-zero.
|
|
|
|
*
|
|
|
|
* @param c pcm channel to examine
|
|
|
|
* @param rates pointer to array of integers; rate table will be recorded here
|
|
|
|
*
|
|
|
|
* @return number of rates in the array pointed to be @c rates
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
chn_getrates(struct pcm_channel *c, int **rates)
|
|
|
|
{
|
|
|
|
KASSERT(rates != NULL, ("rates is null"));
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
return CHANNEL_GETRATES(c->methods, c->devinfo, rates);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @brief Remove channel from a sync group, if there is one.
|
|
|
|
*
|
|
|
|
* This function is initially intended for the following conditions:
|
|
|
|
* - Starting a syncgroup (@c SNDCTL_DSP_SYNCSTART ioctl)
|
|
|
|
* - Closing a device. (A channel can't be destroyed if it's still in use.)
|
|
|
|
*
|
|
|
|
* @note Before calling this function, the syncgroup list mutex must be
|
|
|
|
* held. (Consider pcm_channel::sm protected by the SG list mutex
|
|
|
|
* whether @c c is locked or not.)
|
|
|
|
*
|
|
|
|
* @param c channel device to be started or closed
|
|
|
|
* @returns If this channel was the only member of a group, the group ID
|
|
|
|
* is returned to the caller so that the caller can release it
|
|
|
|
* via free_unr() after giving up the syncgroup lock. Else it
|
|
|
|
* returns 0.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
chn_syncdestroy(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
struct pcmchan_syncmember *sm;
|
|
|
|
struct pcmchan_syncgroup *sg;
|
|
|
|
int sg_id;
|
|
|
|
|
|
|
|
sg_id = 0;
|
|
|
|
|
|
|
|
PCM_SG_LOCKASSERT(MA_OWNED);
|
|
|
|
|
|
|
|
if (c->sm != NULL) {
|
|
|
|
sm = c->sm;
|
|
|
|
sg = sm->parent;
|
|
|
|
c->sm = NULL;
|
|
|
|
|
|
|
|
KASSERT(sg != NULL, ("syncmember has null parent"));
|
|
|
|
|
|
|
|
SLIST_REMOVE(&sg->members, sm, pcmchan_syncmember, link);
|
|
|
|
free(sm, M_DEVBUF);
|
|
|
|
|
|
|
|
if (SLIST_EMPTY(&sg->members)) {
|
|
|
|
SLIST_REMOVE(&snd_pcm_syncgroups, sg, pcmchan_syncgroup, link);
|
|
|
|
sg_id = sg->id;
|
|
|
|
free(sg, M_DEVBUF);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return sg_id;
|
|
|
|
}
|
|
|
|
|
2004-01-28 08:02:15 +00:00
|
|
|
void
|
|
|
|
chn_lock(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
CHN_LOCK(c);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
chn_unlock(struct pcm_channel *c)
|
|
|
|
{
|
|
|
|
CHN_UNLOCK(c);
|
|
|
|
}
|
MFp4 the sound Google Summer of Code project:
The goal was to sync with the OSSv4 API 4Front Technologies uses in their
proprietary OSS driver. This was successful as far as possible. The part
of the API which is stable is implemented, for the rest there are some
stubs already.
New system ioctls:
- SNDCTL_SYSINFO - obtain audio system info (version, # of audio/midi/
mixer devices, etc.)
- SNDCTL_AUDIOINFO - fetch details about a specific audio device
- SNDCTL_MIXERINFO - fetch details about a specific mixer device
New audio ioctls:
- Sync groups (SNDCTL_DSP_SYNCGROUP/SNDCTL_DSP_SYNCSTART) which allow
triggered playback/recording on multiple devices (even across processes
simultaneously).
- Peak meters (SNDCTL_DSP_GETIPEAKS/SNDCTL_DSP_GETOPEAKS) - can query
audio drivers for peak levels (needs driver support, disabled for now).
- Per channel playback/recording levels -
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL. Note that these are still in name
only, just wrapping around the AC97-style mixer at the moment. The next
step is to push them down to the drivers.
Audio ioctls still under development by 4Front (for which stubs may exist
in this commit):
- SNDCTL_GETNAME, SNDCTL_{GET,SET}{SONG,LABEL}
- SNDCTL_DSP_{GET,SET}_CHNORDER
- SNDCTL_MIX_ENUMINFO, SNDCTL_MIX_EXTINFO - (might be documented enough in
the OSS releases to work on this. These ioctls cover the cool "twiddle
any knob on your card" features.)
Missing:
- SNDCTL_DSP_COOKEDMODE -- this ioctl is used to give applications direct
access to a card's buffers, bypassing the feeder architecture. It's
a toughy -- "someone" needs to decide :
(a) if this is desireable, and (b) if it's reasonably feasible.
Updates for driver writers:
So far, only two routines to the channel class (in channel_if.m) are added.
One is for fetching a list of discrete supported playback/recording rates
of a channel, and the other is for fetching peak level info (useful for
drawing peak meters). Interested parties may want to help pushing down
SNDCTL_DSP_{GET,SET}{PLAY,REC}VOL into the drivers.
To use the new stuff you need to rebuild the sound drivers or your kernel
(depending on if you use modules or not) and to install soundcard.h (a
buildworld/installworld handles this).
Sponsored by: Google SoC 2006
Submitted by: ryanb
Many thanks to: 4Front Technologies for their cooperation, explanations
and the nice license of their soundcard.h.
2006-09-23 20:45:47 +00:00
|
|
|
|
|
|
|
#ifdef OSSV4_EXPERIMENT
|
|
|
|
int
|
|
|
|
chn_getpeaks(struct pcm_channel *c, int *lpeak, int *rpeak)
|
|
|
|
{
|
|
|
|
CHN_LOCKASSERT(c);
|
|
|
|
return CHANNEL_GETPEAKS(c->methods, c->devinfo, lpeak, rpeak);
|
|
|
|
}
|
|
|
|
#endif
|