This section answers some of the questions that have been commonly asked on the Usenet news groups and mailing lists.
Answers to more questions can also be found at the OSS sound driver web page.
These are the most "standard" device file names, some Linux distributions may use slightly different names.
normally a link to /dev/audio0
Sun workstation compatible audio device (only a partial implementation, does not support Sun ioctl interface, just u-law encoding)
second audio device (if supported by sound card or if more than one sound card installed)
normally a link to /dev/dsp0
first digital sampling device
second digital sampling device
normally a link to /dev/mixer0
first sound mixer
second sound mixer
high-level sequencer interface
low level MIDI, FM, and GUS access
normally a link to /dev/music
1st raw MIDI port
2nd raw MIDI port
3rd raw MIDI port
4th raw MIDI port
displays sound driver status when read
The PC speaker driver provides the following devices:
equivalent to /dev/audio
equivalent to /dev/dsp
equivalent to /dev/mixer
Sun workstation (.au) sound files can be played by sending them to the
/dev/audio device. Raw samples can be sent to
/dev/dsp. This will generally give poor results though, and
using a program such as play is preferable, as it will
recognize most file types and set the sound card to the correct
sampling rate, etc.
Programs like wavplay or vplay (in the snd-util package) will give best results with WAV files. However they don't recognize Microsoft ADPCM compressed WAV files. Also older versions of play (from the Lsox package) doesn't work well with 16 bit WAV files.
The splay command included in the snd-util package can be used to play most sound files if proper parameters are entered manually in the command line.
Reading /dev/audio or /dev/dsp will return sampled
data that can be redirected to a file. A program such as vrec
makes it easier to control the sampling rate, duration, etc. You may
also need a mixer program to select the appropriate input device.
With the current sound driver it's possible to have several SoundBlaster, SoundBlaster/Pro, SoundBlaster16, MPU-401 or MSS cards at the same time on the system. Installing two SoundBlasters is possible but requires defining the macros SB2_BASE, SB2_IRQ, SB2_DMA and (in some cases) SB2_DMA2 by editing local.h manually. It's also possible to have a SoundBlaster at the same time as a PAS16.
With the newer 2.0.x kernels that configure sound using make config, instead of local.h, you need to edit the file /usr/include/linux/autoconf.h. After the section containing the lines:
#define SBC_BASE 0x220
#define SBC_IRQ (5)
#define SBC_DMA (1)
#define SB_DMA2 (5)
#define SB_MPU_BASE 0x0
#define SB_MPU_IRQ (-1)
add these lines (with values appropriate for your system):
#define SB2_BASE 0x330
#define SB2_IRQ (7)
#define SB2_DMA (2)
#define SB2_DMA2 (2)
The following drivers don't permit multiple instances:
You need to create the sound driver device files. See the section on creating device files. If you do have the device files, ensure that they have the correct major and minor device numbers (some older CD-ROM distributions of Linux may not create the correct device files during installation).
You have not booted with a kernel containing the sound driver or the I/O address configuration doesn't match your hardware. Check that you are running the newly compiled kernel and verify that the settings entered when configuring the sound driver match your hardware setup.
This can happen if you tried to record data to /dev/audio or /dev/dsp without creating the necessary device file. The sound device is now a regular file, and has filled up your disk partition. You need to run the script described in the Creating the Device Files section of this document.
This may also happen with Linux 2.0 and later if there is not enough free RAM on the system when the device is opened. The audio driver requires at least two pages (8k) of contiguous physical RAM for each DMA channel. This happens sometimes in machines with less than 16M of RAM or which have been running for very long time. It may be possible to free some RAM by compiling and running the following C program before trying to open the device again:
main() {
int i;
char mem[500000];
for (i = 0; i < 500000; i++)
mem[i] = 0;
exit(0);
}
Only one process can open a given sound device at one time. Most
likely some other process is using the device in question. One way to
determine this is to use the fuser command:
% fuser -v /dev/dsp
/dev/dsp: USER PID ACCESS COMMAND
tranter 265 f.... tracker
In the above example, the fuser command showed that process
265 had the device open. Waiting for the process to complete or
killing it will allow the sound device to be accessed once again. You
should run the fuser command as root in order to report usage
by users other than yourself.
According to Brian Gough, for the SoundBlaster cards which use DMA channel 1 there is a potential conflict with the QIC-02 tape driver, which also uses DMA 1, causing "device busy" errors. If you are using FTAPE, you may have this driver enabled. According to the FTAPE-HOWTO the QIC-02 driver is not essential for the use of FTAPE; only the QIC-117 driver is required. Reconfiguring the kernel to use QIC-117 but not QIC-02 allows FTAPE and the sound-driver to coexist.
The symptom is usually that a sound sample plays for about a second and then stops completely or reports an error message about "missing IRQ" or "DMA timeout". Most likely you have incorrect IRQ or DMA channel settings. Verify that the kernel configuration matches the sound card jumper settings and that they do not conflict with some other card.
Another symptom is sound samples that "loop". This is usually caused by an IRQ conflict.
Playing MOD files requires considerable CPU power. You may have too many processes running or your computer may be too slow to play in real time. Your options are to:
If you have a Gravis UltraSound card, you should use one of the mod
file players written specifically for the GUS (e.g. gmod).
The version 1.0c and earlier sound driver used a different and
incompatible ioctl() scheme. Obtain newer source code or make
the necessary changes to adapt it to the new sound driver. See the
sound driver Readme file for details.
Also ensure that you have used the latest version of soundcard.h and ultrasound.h when compiling the application. See the installation instructions at beginning of this text.
This is probably the same problem described in the previous question.
See the Readme and CHANGELOG files included with the sound driver kernel source.
These are partially documented in the Hacker's Guide to VoxWare, currently available in draft form. The latest version is draft 2, and can be found on ftp://nic.funet.fi/pub/Linux/ALPHA/sound/. Note that this directory is "hidden" and will not appear in directory listings. If you "cd" to the directory and use the FTP "dir" command, the files are there.
At time of writing new documentation was becoming available on the 4Front Technologies Web site.
Another source of information is the Linux Multimedia Guide, described in the references section.
There is no easy answer to this question, as it depends on:
In general, any 386 machine should be able to play samples or FM synthesized music on an 8 bit sound card with ease.
Playing MOD files, however, requires considerable CPU resources. Some experimental measurements have shown that playing at 44kHz requires more than 40% of the speed of a 486/50 and a 386/25 can hardly play faster than 22 kHz (these are with an 8 bit card sound such as a SoundBlaster). A card such as the Gravis UltraSound card performs more functions in hardware, and will require less CPU resources.
These statements assume the computer is not performing any other CPU intensive tasks.
Converting sound files or adding effects using a utility such as
sox is also much faster if you have a math
coprocessor (or CPU with on board FPU). The kernel driver itself
does not do any floating point calculations, though.
(the following explanation was supplied by
seeker@indirect.com)
Linux only recognizes the 1542 at address 330 (default) or 334, and the PAS only allows the MPU-401 emulation at 330. Even when you disable the MPU-401 under software, something still wants to conflict with the 1542 if it's at its preferred default address. Moving the 1542 to 334 makes everyone happy.
Additionally, both the 1542 and the PAS-16 do 16-bit DMA, so if you sample at 16-bit 44 KHz stereo and save the file to a SCSI drive hung on the 1542, you're about to have trouble. The DMAs overlap and there isn't enough time for RAM refresh, so you get the dread ``PARITY ERROR - SYSTEM HALTED'' message, with no clue to what caused it. It's made worse because a few second-party vendors with QIC-117 tape drives recommend setting the bus on/off times such that the 1542 is on even longer than normal. Get the SCSISEL.EXE program from Adaptec's BBS or several places on the internet, and reduce the BUS ON time or increase the BUS OFF time until the problem goes away, then move it one notch or more further. SCSISEL changes the EEPROM settings, so it's more permanent than a patch to the DOS driver line in CONFIG.SYS, and will work if you boot right into Linux (unlike the DOS patch). Next problem solved.
Last problem - the older Symphony chipsets drastically reduced the timing of the I/O cycles to speed up bus accesses. None of various boards I've played with had any problem with the reduced timing except for the PAS-16. Media Vision's BBS has SYMPFIX.EXE that's supposed to cure the problem by twiddling a diagnostic bit in Symphony's bus controller, but it's not a hard guarantee. You may need to:
Young Microsystems will upgrade the boards they import for around $30 (US); other vendors may be similar if you can figure out who made or imported the motherboard (good luck). The problem is in ProAudio's bus interface chip as far as I'm concerned; nobody buys a $120 sound card and sticks it in a 6MHz AT. Most of them wind up in 25-40MHz 386/486 boxes, and should be able to handle at least 12MHz bus rates if the chips are designed right. Exit soapbox (stage left).
The first problem depends on the chipset used on your motherboard, what bus speed and other BIOS settings, and the phase of the moon. The second problem depends on your refresh option setting (hidden or synchronous), the 1542 DMA rate and (possibly) the bus I/O rate. The third can be determined by calling Media Vision and asking which flavor of Symphony chip is incompatible with their slow design. Be warned, though - 3 of 4 techs I talked to were brain damaged. I would be very leery of trusting anything they said about someone else's hardware, since they didn't even know their own very well.
Due to hardware limitations, this is not possible with most sound cards. Some newer cards do support it. See the section on "bidirectional mode" in the Hacker's Guide to Voxware for more information.
On '286 and later machines, the IRQ 2 interrupt is cascaded to the second interrupt controller. It is equivalent to IRQ 9.
In the past, Creative Labs was not willing to release programming information for these cards. They have since changed their policy and an AWE driver is now included in the Linux 2.1.x kernels.
This happens after a soft reboot to DOS. Sometimes the error message misleadingly refers to a bad CONFIG.SYS file.
Most of the current sound cards have software programmable IRQ and DMA settings. If you use different settings between Linux and MS-DOS/Windows, this may cause problems. Some sound cards don't accept new parameters without a complete reset (i.e. cycle the power or use the hardware reset button).
The quick solution to this problem it to perform a full reboot using the reset button or power cycle rather than a soft reboot (e.g. Ctrl-Alt-Del).
The correct solution is to ensure that you use the same IRQ and DMA settings with MS-DOS and Linux (or not to use DOS :-).
Users of the port of ID software's game DOOM for Linux may be interested in these notes.
For correct sound output you need version 2.90 or later of the sound driver; it has support for the real-time "DOOM mode".
The sound samples are 16-bit. If you have an 8-bit sound card you can still get sound to work using one of several programs available in ftp://sunsite.unc.edu/pub/Linux/games/doom.
If performance of DOOM is poor on your system, disabling sound (by
renaming the file sndserver) may improve it.
By default DOOM does not support music (as in the DOS version). The
program musserver will add support for music to DOOM under
Linux. It can be found at
ftp://pandora.st.hmc.edu/pub/linux/musserver.tgz.
Using good quality shielded cables and trying the sound card in different slots may help reduce noise. If the sound card has a volume control, you can try different settings (maximum is probably best).
Using a mixer program you can make sure that undesired inputs (e.g. microphone) are set to zero gain.
Some sound cards are simply not designed with good shielding and grounding and are prone to noise pickup.
Finally, on my system I found that the kernel command line option
no-hlt reduces the noise level. This tells the kernel not to
use the halt instruction when running the idle process loop. You can
try this manually when booting, or set it up using the command
append="no-hlt" in your LILO configuration file.
If you can play sound but not record, try these steps:
Sometimes a different DMA channel is used for recording than for playback. In this case the most probable reason is that the recording DMA is set up incorrectly.
In most cases a "SoundBlaster compatible" card will work better under Linux if configured with a driver other than the SoundBlaster one. Most sound cards claim to be compatible (e.g. "16 bit SB Pro compatible" or "SB compatible 16 bit") but usually this SoundBlaster mode is just a "hack" provided for DOS games compatibility. Most cards have a 16 bit native mode which is likely to be supported by recent Linux versions (2.0.1 and later).
Only with some (usually rather old) cards is it necessary to try to get them to work in the SoundBlaster mode. The only newer cards that are the exception to this rule are the Mwave-based cards.
16-bit sound cards described as SoundBlaster compatible are really only compatible with the 8-bit SoundBlaster Pro. They typically have a 16-bit mode which is not compatible with the SoundBlaster 16 and not compatible with the Linux sound driver.
You may be able to get the card to work in 16-bit mode by using the MAD16 or MSS/WSS driver.
Here are some good archive sites to search for Linux specific sound applications:
With recent kernels the sound driver is supported as a kernel loadable module.
See the files /usr/src/linux/drivers/sound/Readme.modules and /usr/src/linux/Documentation/modules.txt (or /usr/src/linux/README) for details.
Try the oplbeep program, found at
ftp://sunsite.unc.edu/pub/Linux/apps/sound/oplbeep-alpha.tar.gz
Another variant is the beep program found at
ftp://sunsite.unc.edu/pub/Linux/kernel/patches/misc/modreq_beep.tgz
The modutils package has an example program and kernel patch
that supports calling an arbitrary external program to generate sounds
when requested by the kernel.
Alternatively, with some sound cards you can connect the PC speaker output to the sound card so that all sounds come from the sound card speakers.
The kernel sound drivers support several different Intel-based Unix compatible operating systems, and can be obtained as a package separate from the Linux kernel. Up until February 1996 the author had called the software "VoxWare". Unfortunately this name has been registered by VoxWare Incorporated, and can not be used. The new name of the driver is OSS/Free.
The Open Sound System (OSS) is a commercially available kernel sound driver for various Unix systems, sold by 4Front Technologies. The free version, known as OSS/Free will continue to be made freely available for Linux systems.
Other names you may come across that have been used in the past to refer to the same sound driver are TASD (Temporarily Anonymous Sound Driver) and USS (Unix Sound System).
For more information see the 4Front Technologies Web page at http://www.4front-tech.com/. I wrote a review of OSS/Linux in the June 1997 issue of Linux Journal.
Full Plug and Play support should be coming in Linux version 2.1. In the mean time there are a number of workarounds for getting Plug and Play sound cards to work.
If you have a newer Pentium system with a Plug and Play BIOS, it should take care of configuring the cards for you. Make sure that you configure the Linux sound driver to use the same I/O address, IRQ, and DMA channel parameters as the BIOS.
There is a package of Plug and Play tools for Linux that can be used to set up the card. It can be found at Red Hat's Web site http://www.redhat.com/ (it may also be included in your Linux distribution).
If you use the card under Windows95, you can use the device manager to set up the card, then soft boot into Linux using the LOADLIN program. Make sure Windows95 and Linux use the same card setup parameters.
If you use the card under DOS, you can use the icu utility
that comes with SoundBlaster16 PnP cards to configure it under DOS,
then soft boot into Linux using the LOADLIN program. Again, make sure
DOS and Linux use the same card setup parameters.
The commercial OSS sound driver has support for the SoundBlaster16 PnP sound card. You can purchase this driver from 4Front Technologies.
A change to the sound driver in version 1.3.67 broke some sound player programs which (incorrectly) checked that the result from the SNDCTL_DSP_GETBLKSIZE ioctl was greater than 4096. The utilities included in the latest snd-util-3.x.tar.gz package (at ftp://ftp.4front-tech.com/ossfree.) now handle this properly. The latest sound driver versions have also been fixed to avoid allocating fragments shorter than 4096 bytes which solves this problem with old utilities.
The sound driver supports many different configuration parameters.
The configure program included with the sound driver checks
for many dependencies between parameters. The tools used to configure
the kernel don't support this level of functionality.
That said, the latest kernels do optionally allow using the standard kernel configuration tools with the sound driver (see the earlier section on "Configuring the Kernel".
You can build the sound driver as a loadable module and use
kerneld to automatically load and unload it. This can present
one problem - whenever the module is reloaded the mixer
settings go back to their default values. For some sound cards this
can be too loud (e.g. SoundBlaster16) or too quiet. Markus Gutschke
(gutschk@uni-muenster.de) found this solution. Use a line in
your /etc/conf.modules file such as the following:
options sound dma_buffsize=65536 && /usr/bin/setmixer igain 0 ogain 0 vol 75
This causes your mixer program (in this case setmixer) to be
run immediately after the sound driver is loaded. The
dma_buffsize parameter is just a dummy value needed because
the option command requires a command line option. Change the line as
needed to match your mixer program and gain settings.
If you have compiled the sound driver into your kernel and you want to set the mixer gains at boot time you can put a call to your mixer program in a system startup file such as /etc/rc.d/rc.local.
By default the script in Readme.linux that creates the sound device
files only allows the devices to be read by user root. This
is to plug a potential security hole. In a networked environment,
external users could conceivably log in remotely to a Linux PC with a
sound card and microphone and eavesdrop. If you are not worried about
this, you can change the permissions used in the script.
With the default setup, users can still play sound files. This is not a security risk but is a potential for nuisance.
Information on how to use the mwave sound card on an IBM ThinkPad laptop computer under Linux can be found at http://www.screamin.demon.co.uk/.