freebsd-dev/sys/dev/sound/pcm/channel.c

1686 lines
40 KiB
C
Raw Normal View History

/*-
2003-09-07 16:28:03 +00:00
* Copyright (c) 1999 Cameron Grant <cg@freebsd.org>
* 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.
*/
#include "opt_isa.h"
#include <dev/sound/pcm/sound.h>
#include "feeder_if.h"
SND_DECLARE_FILE("$FreeBSD$");
#define MIN_CHUNK_SIZE 256 /* for uiomove etc. */
#if 0
#define DMA_ALIGN_THRESHOLD 4
#define DMA_ALIGN_MASK (~(DMA_ALIGN_THRESHOLD - 1))
#endif
#define CANCHANGE(c) (!(c->flags & CHN_F_TRIGGERED))
/*
#define DEB(x) x
*/
static int chn_targetirqrate = 32;
TUNABLE_INT("hw.snd.targetirqrate", &chn_targetirqrate);
static int
sysctl_hw_snd_targetirqrate(SYSCTL_HANDLER_ARGS)
{
int err, val;
val = chn_targetirqrate;
err = sysctl_handle_int(oidp, &val, sizeof(val), req);
if (val < 16 || val > 512)
err = EINVAL;
else
chn_targetirqrate = val;
return err;
}
SYSCTL_PROC(_hw_snd, OID_AUTO, targetirqrate, CTLTYPE_INT | CTLFLAG_RW,
0, sizeof(int), sysctl_hw_snd_targetirqrate, "I", "");
static int report_soft_formats = 1;
SYSCTL_INT(_hw_snd, OID_AUTO, report_soft_formats, CTLFLAG_RW,
&report_soft_formats, 1, "report software-emulated formats");
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);
static int chn_buildfeeder(struct pcm_channel *c);
static void
chn_lockinit(struct pcm_channel *c, int dir)
{
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);
}
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);
}
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
*/
static int
chn_polltrigger(struct pcm_channel *c)
{
struct snd_dbuf *bs = c->bufsoft;
unsigned amt, lim;
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
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
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;
}
return 0;
}
static int
chn_pollreset(struct pcm_channel *c)
{
struct snd_dbuf *bs = c->bufsoft;
CHN_LOCKASSERT(c);
sndbuf_updateprevtotal(bs);
return 1;
}
static void
chn_wakeup(struct pcm_channel *c)
{
struct snd_dbuf *bs = c->bufsoft;
struct pcmchan_children *pce;
CHN_LOCKASSERT(c);
if (SLIST_EMPTY(&c->children)) {
if (SEL_WAITING(sndbuf_getsel(bs)) && chn_polltrigger(c))
selwakeuppri(sndbuf_getsel(bs), PRIBIO);
} else {
SLIST_FOREACH(pce, &c->children, link) {
CHN_LOCK(pce->channel);
chn_wakeup(pce->channel);
CHN_UNLOCK(pce->channel);
}
}
wakeup(bs);
}
static int
chn_sleep(struct pcm_channel *c, char *str, int timeout)
{
struct snd_dbuf *bs = c->bufsoft;
int ret;
CHN_LOCKASSERT(c);
#ifdef USING_MUTEX
ret = msleep(bs, c->lock, PRIBIO | PCATCH, str, timeout);
#else
ret = tsleep(bs, PRIBIO | PCATCH, str, timeout);
#endif
return ret;
}
/*
* chn_dmaupdate() tracks the status of a dma transfer,
* updating pointers.
*/
static unsigned int
chn_dmaupdate(struct pcm_channel *c)
{
struct snd_dbuf *b = c->bufhard;
unsigned int delta, old, hwptr, amt;
KASSERT(sndbuf_getsize(b) > 0, ("bufsize == 0"));
CHN_LOCKASSERT(c);
old = sndbuf_gethwptr(b);
hwptr = chn_getptr(c);
delta = (sndbuf_getsize(b) + hwptr - old) % sndbuf_getsize(b);
sndbuf_sethwptr(b, hwptr);
DEB(
if (delta >= ((sndbuf_getsize(b) * 15) / 16)) {
2000-08-09 00:42:00 +00:00
if (!(c->flags & (CHN_F_CLOSING | CHN_F_ABORTING)))
device_printf(c->dev, "hwptr went backwards %d -> %d\n", old, hwptr);
2000-08-09 00:42:00 +00:00
}
);
if (c->direction == PCMDIR_PLAY) {
amt = MIN(delta, sndbuf_getready(b));
if (amt > 0)
sndbuf_dispose(b, NULL, amt);
} else {
amt = MIN(delta, sndbuf_getfree(b));
if (amt > 0)
sndbuf_acquire(b, NULL, amt);
}
return delta;
}
void
chn_wrupdate(struct pcm_channel *c)
{
int ret;
CHN_LOCKASSERT(c);
KASSERT(c->direction == PCMDIR_PLAY, ("chn_wrupdate on bad channel"));
if ((c->flags & (CHN_F_MAPPED | CHN_F_VIRTUAL)) || !(c->flags & CHN_F_TRIGGERED))
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);)
}
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(
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
if (c->flags & CHN_F_MAPPED)
sndbuf_acquire(bs, NULL, sndbuf_getfree(bs));
amt = sndbuf_getfree(b);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
KASSERT(amt <= sndbuf_getsize(bs),
("%s(%s): amt %d > source size %d, flags 0x%x", __func__, c->name,
amt, sndbuf_getsize(bs), c->flags));
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
if (ret == 0 && sndbuf_getfree(b) < amt)
chn_wakeup(c);
return ret;
}
static void
chn_wrintr(struct pcm_channel *c)
{
int ret;
CHN_LOCKASSERT(c);
/* update pointers in primary buffer */
2001-03-25 21:43:24 +00:00
chn_dmaupdate(c);
/* ...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);
DEB(if (ret)
printf("chn_wrintr: chn_wrfeed returned %d\n", ret);)
}
/*
* user write routine - uiomove data into secondary buffer, trigger if necessary
* if blocking, sleep, rinse and repeat.
*
* called externally, so must handle locking
*/
int
chn_write(struct pcm_channel *c, struct uio *buf)
{
int ret, timeout, newsize, count, sz;
struct snd_dbuf *bs = c->bufsoft;
void *off;
int t, x,togo,p;
CHN_LOCKASSERT(c);
/*
* XXX Certain applications attempt to write larger size
* of pcm data than c->blocksize2nd without blocking,
* resulting partial write. Expand the block size so that
* the write operation avoids blocking.
*/
if ((c->flags & CHN_F_NBIO) && buf->uio_resid > sndbuf_getblksz(bs)) {
DEB(device_printf(c->dev, "broken app, nbio and tried to write %d bytes with fragsz %d\n",
buf->uio_resid, sndbuf_getblksz(bs)));
newsize = 16;
while (newsize < min(buf->uio_resid, CHN_2NDBUFMAXSIZE / 2))
newsize <<= 1;
chn_setblocksize(c, sndbuf_getblkcnt(bs), newsize);
DEB(device_printf(c->dev, "frags reset to %d x %d\n", sndbuf_getblkcnt(bs), sndbuf_getblksz(bs)));
}
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
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 {
timeout = (hz * sndbuf_getblksz(bs)) / (sndbuf_getspd(bs) * sndbuf_getbps(bs));
if (timeout < 1)
timeout = 1;
timeout = 1;
ret = chn_sleep(c, "pcmwr", timeout);
if (ret == EWOULDBLOCK) {
count -= timeout;
ret = 0;
} else if (ret == 0)
2001-03-25 21:43:24 +00:00
count = hz;
}
} else {
sz = MIN(sz, buf->uio_resid);
KASSERT(sz > 0, ("confusion in chn_write"));
/* printf("sz: %d\n", sz); */
/*
* The following assumes that the free space in
* the buffer can never be less around the
* unlock-uiomove-lock sequence.
*/
togo = sz;
while (ret == 0 && togo> 0) {
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;
if (ret == 0 && !(c->flags & CHN_F_TRIGGERED))
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;
printf("%s: play interrupt timeout, channel dead\n", c->name);
2001-03-25 21:43:24 +00:00
}
return ret;
}
#if 0
static int
chn_rddump(struct pcm_channel *c, unsigned int cnt)
{
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
static uint32_t kk = 0;
printf("%u: dumping %d bytes\n", ++kk, cnt);
#endif
c->xruns++;
sndbuf_setxrun(b, sndbuf_getxrun(b) + cnt);
return sndbuf_dispose(b, NULL, cnt);
}
#endif
/*
* Feed new data from the read buffer. Can be called in the bottom half.
*/
int
chn_rdfeed(struct pcm_channel *c)
{
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
unsigned int ret, amt;
CHN_LOCKASSERT(c);
DEB(
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
#if 0
amt = sndbuf_getready(b);
2001-10-24 12:33:21 +00:00
if (sndbuf_getfree(bs) < amt) {
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);
ret = (amt > 0)? sndbuf_feed(b, bs, c, c->feeder, amt) : 0;
amt = sndbuf_getready(b);
if (amt > 0) {
c->xruns++;
sndbuf_dispose(b, NULL, amt);
}
chn_wakeup(c);
return ret;
}
void
chn_rdupdate(struct pcm_channel *c)
{
int ret;
CHN_LOCKASSERT(c);
KASSERT(c->direction == PCMDIR_REC, ("chn_rdupdate on bad channel"));
if ((c->flags & CHN_F_MAPPED) || !(c->flags & CHN_F_TRIGGERED))
return;
chn_trigger(c, PCMTRIG_EMLDMARD);
chn_dmaupdate(c);
ret = chn_rdfeed(c);
DEB(if (ret)
printf("chn_rdfeed: %d\n", ret);)
}
/* read interrupt routine. Must be called with interrupts blocked. */
static void
chn_rdintr(struct pcm_channel *c)
{
int ret;
CHN_LOCKASSERT(c);
/* tell the driver to update the primary buffer if non-dma */
2001-03-25 21:43:24 +00:00
chn_trigger(c, PCMTRIG_EMLDMARD);
/* update pointers in primary buffer */
2001-03-25 21:43:24 +00:00
chn_dmaupdate(c);
/* ...and feed from primary to secondary */
ret = chn_rdfeed(c);
}
/*
* user read routine - trigger if necessary, uiomove data from secondary buffer
* if blocking, sleep, rinse and repeat.
*
* called externally, so must handle locking
*/
int
chn_read(struct pcm_channel *c, struct uio *buf)
{
int ret, timeout, sz, count;
struct snd_dbuf *bs = c->bufsoft;
void *off;
int t, x,togo,p;
CHN_LOCKASSERT(c);
if (!(c->flags & CHN_F_TRIGGERED))
chn_start(c, 0);
ret = 0;
2001-03-25 21:43:24 +00:00
count = hz;
while (!ret && (buf->uio_resid > 0) && (count > 0)) {
2001-10-24 12:33:21 +00:00
sz = MIN(buf->uio_resid, sndbuf_getready(bs));
2001-10-24 12:33:21 +00:00
if (sz > 0) {
/*
* The following assumes that the free space in
* the buffer can never be less around the
* unlock-uiomove-lock sequence.
*/
togo = sz;
while (ret == 0 && togo> 0) {
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;
} else {
2001-10-24 12:33:21 +00:00
if (c->flags & CHN_F_NBIO) {
ret = EWOULDBLOCK;
2001-10-24 12:33:21 +00:00
} else {
timeout = (hz * sndbuf_getblksz(bs)) / (sndbuf_getspd(bs) * sndbuf_getbps(bs));
if (timeout < 1)
timeout = 1;
ret = chn_sleep(c, "pcmrd", timeout);
if (ret == EWOULDBLOCK) {
count -= timeout;
ret = 0;
2001-10-24 12:33:21 +00:00
} else {
count = hz;
}
2001-10-24 12:33:21 +00:00
}
2000-06-20 23:42:08 +00:00
}
}
if (count <= 0) {
2001-03-25 21:43:24 +00:00
c->flags |= CHN_F_DEAD;
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
return ret;
}
void
chn_intr(struct pcm_channel *c)
{
CHN_LOCK(c);
c->interrupts++;
2000-04-17 17:06:47 +00:00
if (c->direction == PCMDIR_PLAY)
chn_wrintr(c);
else
chn_rdintr(c);
CHN_UNLOCK(c);
}
u_int32_t
chn_start(struct pcm_channel *c, int force)
{
u_int32_t i, j;
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))
return EINVAL;
i = (c->direction == PCMDIR_PLAY)? sndbuf_getready(bs) : sndbuf_getfree(bs);
j = (c->direction == PCMDIR_PLAY)? sndbuf_getfree(b) : sndbuf_getready(b);
if (force || (i >= j)) {
c->flags |= CHN_F_TRIGGERED;
/*
* if we're starting because a vchan started, don't feed any data
* or it becomes impossible to start vchans synchronised with the
* first one. the hardbuf should be empty so we top it up with
* silence to give it something to chew. the real data will be
* fed at the first irq.
*/
if (c->direction == PCMDIR_PLAY) {
- 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
/*
* Reduce pops during playback startup.
*/
sndbuf_fillsilence(b);
if (SLIST_EMPTY(&c->children))
chn_wrfeed(c);
}
sndbuf_setrun(b, 1);
c->xruns = 0;
chn_trigger(c, PCMTRIG_START);
return 0;
}
return 0;
}
void
chn_resetbuf(struct pcm_channel *c)
{
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
c->blocks = 0;
sndbuf_reset(b);
sndbuf_reset(bs);
}
/*
* chn_sync waits until the space in the given channel goes above
* a threshold. The threshold is checked against fl or rl respectively.
* Assume that the condition can become true, do not check here...
*/
int
chn_sync(struct pcm_channel *c, int threshold)
{
u_long rdy;
int ret;
struct snd_dbuf *bs = c->bufsoft;
CHN_LOCKASSERT(c);
/* 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;
} else {
return 0;
}
}
for (;;) {
rdy = (c->direction == PCMDIR_PLAY)? sndbuf_getfree(bs) : sndbuf_getready(bs);
if (rdy <= threshold) {
ret = chn_sleep(c, "pcmsyn", 1);
if (ret == ERESTART || ret == EINTR) {
DEB(printf("chn_sync: tsleep returns %d\n", ret));
return -1;
}
} else
break;
}
return 0;
}
/* called externally, handle locking */
int
chn_poll(struct pcm_channel *c, int ev, struct thread *td)
{
struct snd_dbuf *bs = c->bufsoft;
int ret;
CHN_LOCKASSERT(c);
if (!(c->flags & CHN_F_MAPPED) && !(c->flags & CHN_F_TRIGGERED))
2001-03-25 21:43:24 +00:00
chn_start(c, 1);
ret = 0;
if (chn_polltrigger(c) && chn_pollreset(c))
ret = ev;
else
selrecord(td, sndbuf_getsel(bs));
return ret;
}
/*
* chn_abort terminates a running dma transfer. it may sleep up to 200ms.
* it returns the number of bytes that have not been transferred.
*
* called from: dsp_close, dsp_ioctl, with channel locked
*/
int
chn_abort(struct pcm_channel *c)
{
int missing = 0;
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
CHN_LOCKASSERT(c);
if (!(c->flags & CHN_F_TRIGGERED))
return 0;
c->flags |= CHN_F_ABORTING;
c->flags &= ~CHN_F_TRIGGERED;
/* kill the channel */
chn_trigger(c, PCMTRIG_ABORT);
sndbuf_setrun(b, 0);
if (!(c->flags & CHN_F_VIRTUAL))
chn_dmaupdate(c);
missing = sndbuf_getready(bs) + sndbuf_getready(b);
c->flags &= ~CHN_F_ABORTING;
2001-03-25 21:43:24 +00:00
return missing;
}
/*
* this routine tries to flush the dma transfer. It is called
* 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.
*
* called from: dsp_close, not valid for record channels.
*/
int
chn_flush(struct pcm_channel *c)
{
int ret, count, resid, resid_p;
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
CHN_LOCKASSERT(c);
KASSERT(c->direction == PCMDIR_PLAY, ("chn_flush on bad channel"));
DEB(printf("chn_flush: c->flags 0x%08x\n", c->flags));
/* 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;
} else {
return 0;
}
}
2001-03-25 21:43:24 +00:00
c->flags |= CHN_F_CLOSING;
resid = sndbuf_getready(bs) + sndbuf_getready(b);
2001-03-25 21:43:24 +00:00
resid_p = resid;
count = 10;
ret = 0;
while ((count > 0) && (resid > sndbuf_getsize(b)) && (ret == 0)) {
2001-03-25 21:43:24 +00:00
/* still pending output data. */
ret = chn_sleep(c, "pcmflu", hz / 10);
if (ret == EWOULDBLOCK)
ret = 0;
if (ret == 0) {
resid = sndbuf_getready(bs) + sndbuf_getready(b);
if (resid == resid_p)
count--;
if (resid > resid_p)
DEB(printf("chn_flush: buffer length increasind %d -> %d\n", resid_p, resid));
resid_p = resid;
2001-03-25 21:43:24 +00:00
}
}
if (count == 0)
DEB(printf("chn_flush: timeout, hw %d, sw %d\n",
sndbuf_getready(b), sndbuf_getready(bs)));
c->flags &= ~CHN_F_TRIGGERED;
/* kill the channel */
chn_trigger(c, PCMTRIG_ABORT);
sndbuf_setrun(b, 0);
c->flags &= ~CHN_F_CLOSING;
return 0;
}
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;
}
int
chn_reset(struct pcm_channel *c, u_int32_t fmt)
{
int hwspd, r;
CHN_LOCKASSERT(c);
c->flags &= CHN_F_RESET;
c->interrupts = 0;
c->xruns = 0;
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
hwspd = DSP_DEFAULT_SPEED;
/* 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);
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;
if (r == 0)
r = chn_setformat(c, fmt);
if (r == 0)
r = chn_setspeed(c, hwspd);
#if 0
if (r == 0)
r = chn_setvolume(c, 100, 100);
#endif
2000-04-17 17:06:47 +00:00
}
if (r == 0)
r = chn_setblocksize(c, 0, 0);
if (r == 0) {
2001-03-25 21:43:24 +00:00
chn_resetbuf(c);
r = CHANNEL_RESETDONE(c->methods, c->devinfo);
}
return r;
}
int
chn_init(struct pcm_channel *c, void *devinfo, int dir, int direction)
{
struct feeder_class *fc;
struct snd_dbuf *b, *bs;
int ret;
chn_lockinit(c, dir);
b = NULL;
bs = NULL;
c->devinfo = NULL;
c->feeder = NULL;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
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);
ret = EINVAL;
fc = feeder_getclass(NULL);
if (fc == NULL)
goto out;
if (chn_addfeeder(c, fc, NULL))
goto out;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
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);
sndbuf_setup(bs, NULL, 0);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(c);
c->bufhard = b;
c->bufsoft = bs;
c->flags = 0;
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;
ret = ENODEV;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_UNLOCK(c); /* XXX - Unlock for CHANNEL_INIT() malloc() call */
c->devinfo = CHANNEL_INIT(c->methods, devinfo, b, c, direction);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(c);
if (c->devinfo == NULL)
goto out;
ret = ENOMEM;
if ((sndbuf_getsize(b) == 0) && ((c->flags & CHN_F_VIRTUAL) == 0))
goto out;
ret = chn_setdir(c, direction);
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;
}
}
out:
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_UNLOCK(c);
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;
}
return 0;
}
int
chn_kill(struct pcm_channel *c)
{
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
if (c->flags & CHN_F_TRIGGERED)
chn_trigger(c, PCMTRIG_ABORT);
while (chn_removefeeder(c) == 0);
if (CHANNEL_FREE(c->methods, c->devinfo))
sndbuf_free(b);
c->flags |= CHN_F_DEAD;
sndbuf_destroy(bs);
sndbuf_destroy(b);
chn_lockdestroy(c);
return 0;
}
int
chn_setdir(struct pcm_channel *c, int dir)
{
#ifdef DEV_ISA
struct snd_dbuf *b = c->bufhard;
#endif
int r;
CHN_LOCKASSERT(c);
c->direction = dir;
r = CHANNEL_SETDIR(c->methods, c->devinfo, c->direction);
#ifdef DEV_ISA
if (!r && SND_DMA(b))
sndbuf_dmasetdir(b, c->direction);
#endif
return r;
}
int
chn_setvolume(struct pcm_channel *c, int left, int right)
{
CHN_LOCKASSERT(c);
/* should add a feeder for volume changing if channel returns -1 */
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;
}
static int
chn_tryspeed(struct pcm_channel *c, int speed)
{
struct pcm_feeder *f;
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
struct snd_dbuf *x;
int r, delta;
CHN_LOCKASSERT(c);
DEB(printf("setspeed, channel %s\n", c->name));
DEB(printf("want speed %d, ", speed));
if (speed <= 0)
return EINVAL;
if (CANCHANGE(c)) {
r = 0;
c->speed = speed;
sndbuf_setspd(bs, speed);
RANGE(speed, chn_getcaps(c)->minspeed, chn_getcaps(c)->maxspeed);
DEB(printf("try speed %d, ", speed));
sndbuf_setspd(b, CHANNEL_SETSPEED(c->methods, c->devinfo, speed));
DEB(printf("got speed %d\n", sndbuf_getspd(b)));
delta = sndbuf_getspd(b) - sndbuf_getspd(bs);
if (delta < 0)
delta = -delta;
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!
*/
if (delta > 25)
c->feederflags |= 1 << FEEDER_RATE;
else
sndbuf_setspd(bs, sndbuf_getspd(b));
r = chn_buildfeeder(c);
DEB(printf("r = %d\n", r));
if (r)
goto out;
r = chn_setblocksize(c, 0, 0);
if (r)
goto 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 (!(c->feederflags & (1 << FEEDER_RATE)))
goto out;
r = EINVAL;
f = chn_findfeeder(c, FEEDER_RATE);
DEB(printf("feedrate = %p\n", f));
if (f == NULL)
goto out;
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));
if (r)
goto out;
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));
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);
DEB(printf("setspeed done, r = %d\n", r));
2001-03-25 21:43:24 +00:00
return r;
} else
return EINVAL;
}
int
chn_setspeed(struct pcm_channel *c, int speed)
{
int r, oldspeed = c->speed;
r = chn_tryspeed(c, speed);
if (r) {
DEB(printf("Failed to set speed %d falling back to %d\n", speed, oldspeed));
r = chn_tryspeed(c, oldspeed);
}
return r;
}
static int
chn_tryformat(struct pcm_channel *c, u_int32_t fmt)
{
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
int r;
CHN_LOCKASSERT(c);
if (CANCHANGE(c)) {
DEB(printf("want format %d\n", fmt));
c->format = fmt;
r = chn_buildfeeder(c);
if (r == 0) {
sndbuf_setfmt(bs, c->format);
2001-03-25 21:43:24 +00:00
chn_resetbuf(c);
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
}
return r;
} else
return EINVAL;
}
int
chn_setformat(struct pcm_channel *c, u_int32_t fmt)
{
u_int32_t oldfmt = c->format;
int r;
r = chn_tryformat(c, fmt);
if (r) {
DEB(printf("Format change %d failed, reverting to %d\n", fmt, oldfmt));
chn_tryformat(c, oldfmt);
}
return r;
}
/*
* given a bufsz value, round it to a power of 2 in the min-max range
* XXX only works if min and max are powers of 2
*/
static int
round_bufsz(int bufsz, int min, int max)
{
int tmp = min * 2;
KASSERT((min & (min-1)) == 0, ("min %d must be power of 2\n", min));
KASSERT((max & (max-1)) == 0, ("max %d must be power of 2\n", max));
while (tmp <= bufsz)
tmp <<= 1;
tmp >>= 1;
if (tmp > max)
tmp = max;
return tmp;
}
/*
* set the channel's blocksize both for soft and hard buffers.
*
* blksz should be a power of 2 between 2**4 and 2**16 -- it is useful
* that it has the same value for both bufsoft and bufhard.
* blksz == -1 computes values according to a target irq rate.
* blksz == 0 reuses previous values if available, otherwise
* behaves as for -1
*
* blkcnt is set by the user, between 2 and (2**17)/blksz for bufsoft,
* but should be a power of 2 for bufhard to simplify life to low
* level drivers.
* Note, for the rec channel a large blkcnt is ok,
* but for the play channel we want blksz as small as possible to keep
* the delay small, because routines in the write path always try to
* keep bufhard full.
*
* Unless we have good reason to, use the values suggested by the caller.
*/
int
chn_setblocksize(struct pcm_channel *c, int blkcnt, int blksz)
{
struct snd_dbuf *b = c->bufhard;
struct snd_dbuf *bs = c->bufsoft;
int irqhz, ret, maxsz, maxsize, reqblksz;
CHN_LOCKASSERT(c);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
if (!CANCHANGE(c) || (c->flags & CHN_F_MAPPED)) {
KASSERT(sndbuf_getsize(bs) == 0 ||
sndbuf_getsize(bs) >= sndbuf_getsize(b),
("%s(%s): bufsoft size %d < bufhard size %d", __func__,
c->name, sndbuf_getsize(bs), sndbuf_getsize(b)));
return EINVAL;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
}
c->flags |= CHN_F_SETBLOCKSIZE;
ret = 0;
DEB(printf("%s(%d, %d)\n", __func__, blkcnt, blksz));
if (blksz == 0 || blksz == -1) { /* let the driver choose values */
if (blksz == -1) /* delete previous values */
c->flags &= ~CHN_F_HAS_SIZE;
if (!(c->flags & CHN_F_HAS_SIZE)) { /* no previous value */
/*
* compute a base blksz according to the target irq
* rate, then round to a suitable power of 2
* in the range 16.. 2^17/2.
* Finally compute a suitable blkcnt.
*/
blksz = round_bufsz( (sndbuf_getbps(bs) *
sndbuf_getspd(bs)) / chn_targetirqrate,
16, CHN_2NDBUFMAXSIZE / 2);
blkcnt = CHN_2NDBUFMAXSIZE / blksz;
} else { /* use previously defined value */
blkcnt = sndbuf_getblkcnt(bs);
blksz = sndbuf_getblksz(bs);
}
} else {
/*
* use supplied values if reasonable. Note that here we
* might have blksz which is not a power of 2 if the
* ioctl() to compute it allows such values.
*/
ret = EINVAL;
if ((blksz < 16) || (blkcnt < 2) || (blkcnt * blksz > CHN_2NDBUFMAXSIZE))
goto out;
ret = 0;
c->flags |= CHN_F_HAS_SIZE;
}
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
reqblksz = blksz;
- 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 (reqblksz < sndbuf_getbps(bs))
reqblksz = sndbuf_getbps(bs);
if (reqblksz % sndbuf_getbps(bs))
reqblksz -= reqblksz % sndbuf_getbps(bs);
/* adjust for different hw format/speed */
/*
* Now compute the approx irq rate for the given (soft) blksz,
* reduce to the acceptable range and compute a corresponding blksz
* for the hard buffer. Then set the channel's blocksize and
* corresponding hardbuf value. The number of blocks used should
* be set by the device-specific routine. In fact, even the
* call to sndbuf_setblksz() should not be here! XXX
*/
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
irqhz = (sndbuf_getbps(bs) * sndbuf_getspd(bs)) / blksz;
RANGE(irqhz, 16, 512);
maxsz = sndbuf_getmaxsize(b);
if (maxsz == 0) /* virtual channels don't appear to allocate bufhard */
maxsz = CHN_2NDBUFMAXSIZE;
blksz = round_bufsz( (sndbuf_getbps(b) * sndbuf_getspd(b)) / irqhz,
16, maxsz / 2);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
/* Increase the size of bufsoft if before increasing bufhard. */
maxsize = sndbuf_getsize(b);
if (sndbuf_getsize(bs) > maxsize)
maxsize = sndbuf_getsize(bs);
if (reqblksz * blkcnt > maxsize)
maxsize = reqblksz * blkcnt;
if (sndbuf_getsize(bs) != maxsize || sndbuf_getblksz(bs) != reqblksz) {
ret = sndbuf_remalloc(bs, maxsize/reqblksz, reqblksz);
if (ret)
goto out1;
}
CHN_UNLOCK(c);
sndbuf_setblksz(b, CHANNEL_SETBLOCKSIZE(c->methods, c->devinfo, blksz));
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(c);
/* Decrease the size of bufsoft after decreasing bufhard. */
maxsize = sndbuf_getsize(b);
if (reqblksz * blkcnt > maxsize)
maxsize = reqblksz * blkcnt;
if (maxsize > sndbuf_getsize(bs))
printf("Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE()\n",
c->name, sndbuf_getsize(bs), maxsize);
if (sndbuf_getsize(bs) != maxsize || sndbuf_getblksz(bs) != reqblksz) {
ret = sndbuf_remalloc(bs, maxsize/reqblksz, reqblksz);
if (ret)
goto out1;
}
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
/*
* 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);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
out1:
KASSERT(sndbuf_getsize(bs) == 0 ||
sndbuf_getsize(bs) >= sndbuf_getsize(b),
("%s(%s): bufsoft size %d < bufhard size %d, reqblksz=%d blksz=%d maxsize=%d blkcnt=%d",
__func__, c->name, sndbuf_getsize(bs), sndbuf_getsize(b), reqblksz,
blksz, maxsize, blkcnt));
out:
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
c->flags &= ~CHN_F_SETBLOCKSIZE;
- 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
if (1) {
static uint32_t kk = 0;
printf("%u: b %d/%d/%d : (%d)%d/0x%0x | bs %d/%d/%d : (%d)%d/0x%0x\n", ++kk,
sndbuf_getsize(b), sndbuf_getblksz(b), sndbuf_getblkcnt(b),
sndbuf_getbps(b),
sndbuf_getspd(b), sndbuf_getfmt(b),
sndbuf_getsize(bs), sndbuf_getblksz(bs), sndbuf_getblkcnt(bs),
sndbuf_getbps(bs),
sndbuf_getspd(bs), sndbuf_getfmt(bs));
if (sndbuf_getsize(b) % sndbuf_getbps(b) ||
sndbuf_getblksz(b) % sndbuf_getbps(b) ||
sndbuf_getsize(bs) % sndbuf_getbps(bs) ||
sndbuf_getblksz(b) % sndbuf_getbps(b)) {
printf("%u: bps/blksz alignment screwed!\n", kk);
}
}
#endif
return ret;
}
int
chn_trigger(struct pcm_channel *c, int go)
{
#ifdef DEV_ISA
struct snd_dbuf *b = c->bufhard;
#endif
int ret;
CHN_LOCKASSERT(c);
#ifdef DEV_ISA
if (SND_DMA(b) && (go == PCMTRIG_EMLDMAWR || go == PCMTRIG_EMLDMARD))
sndbuf_dmabounce(b);
#endif
ret = CHANNEL_TRIGGER(c->methods, c->devinfo, go);
return ret;
}
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
*/
int
chn_getptr(struct pcm_channel *c)
{
#if 0
int hwptr;
int a = (1 << c->align) - 1;
CHN_LOCKASSERT(c);
hwptr = (c->flags & CHN_F_TRIGGERED)? CHANNEL_GETPTR(c->methods, c->devinfo) : 0;
/* don't allow unaligned values in the hwa ptr */
#if 1
hwptr &= ~a ; /* Apply channel align mask */
#endif
hwptr &= DMA_ALIGN_MASK; /* Apply DMA align mask */
return hwptr;
#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)));
}
struct pcmchan_caps *
chn_getcaps(struct pcm_channel *c)
{
CHN_LOCKASSERT(c);
return CHANNEL_GETCAPS(c->methods, c->devinfo);
}
u_int32_t
chn_getformats(struct pcm_channel *c)
{
u_int32_t *fmtlist, fmts;
int i;
fmtlist = chn_getcaps(c)->fmtlist;
fmts = 0;
for (i = 0; fmtlist[i]; i++)
fmts |= fmtlist[i];
/* 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|
AFMT_S16_LE|AFMT_S16_BE|AFMT_U8|AFMT_S8;
return fmts;
}
static int
chn_buildfeeder(struct pcm_channel *c)
{
struct feeder_class *fc;
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;
int err;
CHN_LOCKASSERT(c);
while (chn_removefeeder(c) == 0);
KASSERT((c->feeder == NULL), ("feeder chain not empty"));
c->align = sndbuf_getalign(c->bufsoft);
if (SLIST_EMPTY(&c->children)) {
fc = feeder_getclass(NULL);
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;
}
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;
desc.in = 0;
} else {
DEB(printf("can't decide which feeder type to use!\n"));
return EOPNOTSUPP;
}
desc.out = c->format;
desc.flags = 0;
fc = feeder_getclass(&desc);
if (fc == NULL) {
DEB(printf("can't find vchan feeder\n"));
return EOPNOTSUPP;
}
err = chn_addfeeder(c, fc, &desc);
if (err) {
DEB(printf("can't add vchan feeder, err %d\n", err));
return err;
}
}
c->feederflags &= ~(1 << FEEDER_VOLUME);
if (c->direction == PCMDIR_PLAY &&
!(c->flags & CHN_F_VIRTUAL) &&
c->parentsnddev && (c->parentsnddev->flags & SD_F_SOFTPCMVOL) &&
c->parentsnddev->mixer_dev)
c->feederflags |= 1 << FEEDER_VOLUME;
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;
DEB(printf("feederflags %x\n", flags));
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));
fc = feeder_getclass(&desc);
DEB(printf("got %p\n", fc));
if (fc == NULL) {
DEB(printf("can't find required feeder type %d\n", type));
return EOPNOTSUPP;
}
DEB(printf("build fmtchain from 0x%08x to 0x%08x: ", c->feeder->desc->out, fc->desc->in));
tmp[0] = fc->desc->in;
tmp[1] = 0;
if (chn_fmtchain(c, tmp) == 0) {
DEB(printf("failed\n"));
return ENODEV;
}
DEB(printf("ok\n"));
err = chn_addfeeder(c, fc, fc->desc);
if (err) {
DEB(printf("can't add feeder %p, output 0x%x, err %d\n", fc, fc->desc->out, err));
return err;
}
DEB(printf("added feeder %p, output 0x%x\n", fc, c->feeder->desc->out));
}
}
if (c->direction == PCMDIR_REC) {
tmp[0] = c->format;
tmp[1] = 0;
hwfmt = chn_fmtchain(c, tmp);
} else
hwfmt = chn_fmtchain(c, fmtlist);
- 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)) {
DEB(printf("Invalid hardware format: 0x%08x\n", hwfmt));
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
}
sndbuf_setfmt(c->bufhard, hwfmt);
if ((flags & (1 << FEEDER_VOLUME))) {
uint32_t parent = SOUND_MIXER_NONE;
int vol, left, right;
vol = 100 | (100 << 8);
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)
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;
}
CHN_LOCK(c);
chn_setvolume(c, left, right);
}
return 0;
}
int
chn_notify(struct pcm_channel *c, u_int32_t flags)
{
struct pcmchan_children *pce;
struct pcm_channel *child;
int run;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(c);
if (SLIST_EMPTY(&c->children)) {
CHN_UNLOCK(c);
return ENODEV;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +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;
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) {
int blksz;
/*
* scan the children, find the lowest blocksize and use that
* for the hard blocksize
*/
blksz = sndbuf_getmaxsize(c->bufhard) / 2;
SLIST_FOREACH(pce, &c->children, link) {
child = pce->channel;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(child);
if (sndbuf_getblksz(child->bufhard) < blksz)
blksz = sndbuf_getblksz(child->bufhard);
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_UNLOCK(child);
}
chn_setblocksize(c, 2, blksz);
}
if (flags & CHN_N_TRIGGER) {
int nrun;
/*
* 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
*/
nrun = 0;
SLIST_FOREACH(pce, &c->children, link) {
child = pce->channel;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_LOCK(child);
if (child->flags & CHN_F_TRIGGERED)
nrun = 1;
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_UNLOCK(child);
}
if (nrun && !run)
chn_start(c, 1);
if (!nrun && run)
chn_abort(c);
}
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
2004-01-28 08:02:15 +00:00
CHN_UNLOCK(c);
return 0;
}
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
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;
}
Change KASSERT() in feed_vchan16() into an explicit test and call to panic() so that the buffer overflow just beyond this point is always caught, even when the code is not compiled with INVARIANTS. Change chn_setblocksize() buffer reallocation code to attempt to avoid the feed_vchan16() buffer overflow by attempting to always keep the bufsoft buffer at least as large as the bufhard buffer. Print a diagnositic message Danger! %s bufsoft size increasing from %d to %d after CHANNEL_SETBLOCKSIZE() if our best attempts fail. If feed_vchan16() were to be called by the interrupt handler while locks are dropped in chn_setblocksize() to increase the size bufsoft to match the size of bufhard, the panic() code in feed_vchan16() will be triggered. If the diagnostic message is printed, it is a warning that a panic is possible if the system were to see events in an "unlucky" order. Change the locking code to avoid the need for MTX_RECURSIVE mutexes. Add the MTX_DUPOK option to the channel mutexes and change the locking sequence to always lock the parent channel before its children to avoid the possibility of deadlock. Actually implement locking assertions for the channel mutexes and fix the problems found by the resulting assertion violations. Clean up the locking code in dsp_ioctl(). Allocate the channel buffers using the malloc() M_WAITOK option instead of M_NOWAIT so that buffer allocation won't fail. Drop locks across the malloc() calls. Add/modify KASSERTS() in attempt to detect problems early. Abuse layering by adding a pointer to the snd_dbuf structure that points back to the pcm_channel that owns it. This allows sndbuf_resize() to do proper locking without having to change the its API, which is used by the hardware drivers. Don't dereference a NULL pointer when setting hw.snd.maxautovchans if a hardware driver is not loaded. Noticed by Ryan Sommers <ryans at gamersimpact.com>. Tested by: Stefan Ehmann <shoesoft AT gmx.net> Tested by: matk (Mathew Kanner) Tested by: Gordon Bergling <gbergling AT 0xfce3.net>
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