SOUND(4)

📦 version 12.2, in debian bookworm, debian trixie, ubuntu jammy, ubuntu noble, ubuntu questing, ubuntu resolute, ubuntu stonking

💖 SOUND(4) 💖

okay bestie, you clicked on the sound, pcm, snd page — FreeBSD PCM audio device infrastructure — and honestly? it's giving main character energy for your whole sound card situation. this is section 4 of the Kernel Interfaces Manual, and it's a long one, so buckle up 🎀

✨ NAME ✨

sound, pcm, snd — FreeBSD PCM audio device infrastructure

🦋 SYNOPSIS 🦋

so to compile this driver into the kernel, you drop this line in your kernel configuration file:

device sound

periodt. that's the whole vibe.

👑 DESCRIPTION 👑

The sound driver is the main component of the FreeBSD sound system, no cap. It works in conjunction with a bridge device driver on supported devices, and once it attaches it provides PCM audio record and playback. Each bridge device driver supports its own specific set of audio chipsets, and it needs to be enabled together with the sound driver. PCI and ISA PnP audio devices identify themselves, so bestie you usually do not have to add anything to /boot/device.hints. iconic behavior honestly.

Some of the main features of the sound driver are, and I'm obsessed:

  • multichannel audio
  • per-application volume control
  • dynamic mixing through virtual sound channels
  • true full duplex operation
  • bit perfect audio
  • rate conversion
  • low latency modes

The sound driver is enabled by default, along with several bridge device drivers. The ones that are not enabled by default can be loaded during runtime with kldload(8) or during boot via loader.conf(5). Here are all the bridge device drivers available — every single one, bestie, I'm not skipping a single one 💅:

Refer to the manual page for each bridge device driver for driver specific settings and information. iykyk 🌸

🌈 Legacy Hardware 🌈

For old legacy ISA cards, the driver looks for MSS cards at addresses 0x530 and 0x604. These values can be overridden in /boot/device.hints. Non-PnP sound cards require the following lines in device.hints(5):

hint.pcm.0.at="isa"
hint.pcm.0.irq="5"
hint.pcm.0.drq="1"
hint.pcm.0.flags="0x0"

Apart from the usual parameters, the flags field is used to specify the secondary DMA channel (generally used for capture in full duplex cards). Flags are set to 0 for cards not using a secondary DMA channel, or to 0x10 + C to specify channel C. yes bestie, math class snuck in 🙈

🦋 Boot Variables 🦋

In general, the module snd_foo corresponds to device snd_foo and can be loaded by the boot loader(8) via loader.conf(5) or from the command line using the kldload(8) utility. Options which can be specified in /boot/loader.conf include:

  • snd_driver_load (“NO”) — If set to “YES”, this option loads all available drivers.
  • snd_hda_load (“NO”) — If set to “YES”, only the Intel High Definition Audio bridge device driver and dependent modules will be loaded.
  • snd_foo_load (“NO”) — If set to “YES”, load driver for card/chipset foo.

To define default values for the different mixer channels, set the channel to the preferred value using hints, e.g.: hint.pcm.0.line="0". This will mute the input channel per default. 💀

💅 Multichannel Audio 💅

Multichannel audio, popularly referred to as “surround sound”, is supported and enabled by default. The FreeBSD multichannel matrix processor supports up to 18 interleaved channels, but the limit is currently set to 8 channels (as commonly used for 7.1 surround sound). The internal matrix mapping can handle reduction, expansion or re-routing of channels. This provides a base interface for related multichannel ioctl() support. Multichannel audio works both with and without VCHANs. slay.

Most bridge device drivers are still missing multichannel matrixing support, but in most cases this should be trivial to implement. Use the dev.pcm.%d.[play|rec].vchanformat sysctl(8) to adjust the number of channels used. The current multichannel interleaved structure and arrangement was implemented by inspecting various popular UNIX applications. There was no single standard, so much care has been taken to try to satisfy each possible scenario, despite the fact that each application has its own conflicting standard. not the conflicting standards 😭

⭐ EQ ⭐

The Parametric Software Equalizer (EQ) enables the use of “tone” controls (bass and treble). Commonly used for ear-candy or frequency compensation due to the vast difference in hardware quality. EQ is disabled by default, but can be enabled with the hint.pcm.%d.eq tunable.

🎀 VCHANs 🎀

Each device can optionally support more playback and recording channels than physical hardware provides by using “virtual channels” or VCHANs. VCHAN options can be configured via the sysctl(8) interface but can only be manipulated while the device is inactive. important!! don't try to do it while it's busy bestie 🫶

🔥 VPC 🔥

FreeBSD supports independent and individual volume controls for each active application, without touching the master sound volume. This is sometimes referred to as Volume Per Channel (VPC). The VPC feature is enabled by default. per-app volume control?? we love to see it 👑

💜 Loader Tunables 💜

The following loader tunables are used to set driver configuration at the loader(8) prompt before booting the kernel, or they can be stored in /boot/loader.conf in order to automatically set them before booting the kernel. It is also possible to use kenv(1) to change these tunables before loading the sound driver. These following tunables can not be changed during runtime using sysctl(8). so set them early or cry about it later 💀

  • hint.pcm.%d.eq — Set to 1 or 0 to explicitly enable (1) or disable (0) the equalizer. Requires a driver reload if changed. Enabling this will make bass and treble controls appear in mixer applications. This tunable is undefined by default. Equalizing is disabled by default.
  • hint.pcm.%d.vpc — Set to 1 or 0 to explicitly enable (1) or disable (0) the VPC feature. This tunable is undefined by default. VPC is however enabled by default.

🌸 Runtime Configuration 🌸

There are a number of sysctl(8) variables available which can be modified during runtime. These values can also be stored in /etc/sysctl.conf in order to automatically set them during the boot process. hw.snd.* are global settings and dev.pcm.* are device specific. 🧚

  • hw.snd.compat_linux_mmap — Linux mmap(2) compatibility. The following values are supported (default is 0):
    • -1 — Force disabling/denying PROT_EXEC mmap(2) requests.
    • 0 — Auto detect proc/ABI type, allow mmap(2) for Linux applications, and deny for everything else.
    • 1 — Always allow PROT_EXEC page mappings.
  • hw.snd.default_auto — Automatically assign the default sound unit. The following values are supported (default is 1):
    • 0 — Do not assign the default sound unit automatically.
    • 1 — Use the best available sound device based on playing and recording capabilities of the device.
    • 2 — Use the most recently attached device.
  • hw.snd.default_unit — Default sound card for systems with multiple sound cards. When using devfs(5), the default device for /dev/dsp. Equivalent to a symlink from /dev/dsp to /dev/dsp${hw.snd.default_unit}.
  • hw.snd.feeder_eq_exact_rate — Only certain rates are allowed for precise processing. The default behavior is however to allow sloppy processing for all rates, even the unsupported ones. Enable to toggle this requirement and only allow processing for supported rates.
  • hw.snd.feeder_rate_max — Maximum allowable sample rate.
  • hw.snd.feeder_rate_min — Minimum allowable sample rate.
  • hw.snd.feeder_rate_polyphase_max — Adjust to set the maximum number of allowed polyphase entries during the process of building resampling filters. Disabling polyphase resampling has the benefit of reducing memory usage, at the expense of slower and lower quality conversion. Only applicable when the SINC interpolator is used. Default value is 183040. Set to 0 to disable polyphase resampling.
  • hw.snd.feeder_rate_quality — Sample rate converter quality. Default value is 1, linear interpolation. Available options include:
    • 0 — Zero Order Hold, ZOH. Very fast, but with poor quality.
    • 1 — Linear interpolation. Fast, quality is subject to personal preference. Technically the quality is poor however, due to the lack of anti-aliasing filtering.
    • 2 — Bandlimited SINC interpolator. Implements polyphase banking to boost the conversion speed, at the cost of memory usage, with multiple high quality polynomial interpolators to improve the conversion accuracy. 100% fixed point, 64bit accumulator with 32bit coefficients and high precision sample buffering. Quality values are 100dB stopband, 8 taps and 85% bandwidth.
    • 3 — Continuation of the bandlimited SINC interpolator, with 100dB stopband, 36 taps and 90% bandwidth as quality values.
    • 4 — Continuation of the bandlimited SINC interprolator, with 100dB stopband, 164 taps and 97% bandwidth as quality values.
  • hw.snd.feeder_rate_round — Sample rate rounding threshold, to avoid large prime division at the cost of accuracy. All requested sample rates will be rounded to the nearest threshold value. Possible values range between 0 (disabled) and 500. Default is 25.
  • hw.snd.latency — Configure the buffering latency. Only affects applications that do not explicitly request blocksize / fragments. This tunable provides finer granularity than the hw.snd.latency_profile tunable. Possible values range between 0 (lowest latency) and 10 (highest latency).
  • hw.snd.latency_profile — bestie this one defines sets of buffering latency conversion tables for the hw.snd.latency tunable. A value of 0 will use a low and aggressive latency profile which can result in possible underruns if the application cannot keep up with a rapid irq rate, especially during high workload. The default value is 1, which is considered a moderate/safe latency profile. ✨
  • hw.snd.maxautovchans — global VCHAN setting that only affects devices with at least one playback or recording channel available. The sound system will dynamically create up to this many VCHANs. Set to “0” if no VCHANs are desired. Maximum value is 256. 👑
  • hw.snd.report_soft_formats — controls the internal format conversion if it is available transparently to the application software. When disabled or not available, the application will only be able to select formats the device natively supports. 💅
  • hw.snd.report_soft_matrix — enable seamless channel matrixing even if the hardware does not support it. Makes it possible to play multichannel streams even with a simple stereo sound card. 🌈
  • hw.snd.verbose — level of verbosity for the /dev/sndstat device. Higher values include more output and the highest level, four, should be used when reporting problems. Other options include:
    • 0 — Installed devices and their allocated bus resources.
    • 1 — The number of playback, record, virtual channels, and flags per device.
    • 2 — Channel information per device including the channel's current format, speed, and pseudo device statistics such as buffer overruns and buffer underruns.
    • 3 — File names and versions of the currently loaded sound modules.
    • 4 — Various messages intended for debugging.
  • hw.snd.vpc_0db — default value for sound volume. Increase to give more room for attenuation control. Decrease for more amplification, with the possible cost of sound clipping. 😭
  • hw.snd.vpc_autoreset — when a channel is closed the channel volume will be reset to 0db. This means that any changes to the volume will be lost. Enabling this will preserve the volume, at the cost of possible confusion when applications tries to re-open the same device. 🫶
  • hw.snd.vpc_mixer_bypass — the recommended way to use the VPC feature is to teach applications to use the correct ioctl(): SNDCTL_DSP_GETPLAYVOL, SNDCTL_DSP_SETPLAYVOL, SNDCTL_DSP_SETRECVOL, SNDCTL_DSP_SETRECVOL. This is however not always possible. Enable this to allow applications to use their own existing mixer logic to control their own channel volume. 💀
  • hw.snd.vpc_reset — enable to restore all channel volumes back to the default value of 0db. 🎀
  • dev.pcm.%d.bitperfect — enable or disable bitperfect mode. When enabled, channels will skip all dsp processing, such as channel matrixing, rate converting and equalizing. The pure sound stream will be fed directly to the hardware. If VCHANs are enabled, the bitperfect mode will use the VCHAN format/rate as the definitive format/rate target. The recommended way to use bitperfect mode is to disable VCHANs and enable this sysctl. Default is disabled. 🔥
  • dev.pcm.%d.[play|rec].vchans — the current number of VCHANs allocated per device. This can be set to preallocate a certain number of VCHANs. Setting this value to “0” will disable VCHANs for this device. ✨
  • dev.pcm.%d.[play|rec].vchanformat — format for VCHAN mixing. All playback paths will be converted to this format before the mixing process begins. By default only 2 channels are enabled. Available options include:
    • s16le:1.0 — Mono.
    • s16le:2.0 — Stereo, 2 channels (left, right).
    • s16le:2.1 — 3 channels (left, right, LFE).
    • s16le:3.0 — 3 channels (left, right, rear center).
    • s16le:4.0 — Quadraphonic, 4 channels (front/rear left and right).
    • s16le:4.1 — 5 channels (4.0 + LFE).
    • s16le:5.0 — 5 channels (4.0 + center).
    • s16le:5.1 — 6 channels (4.0 + center + LFE).
    • s16le:6.0 — 6 channels (4.0 + front/rear center).
    • s16le:6.1 — 7 channels (6.0 + LFE).
    • s16le:7.1 — 8 channels (4.0 + center + LFE + left and right side).
  • dev.pcm.%d.[play|rec].vchanmode — VCHAN format/rate selection. Available options include:
    • fixed — Channel mixing is done using fixed format/rate. Advanced operations such as digital passthrough will not work. Can be considered as a “legacy” mode. This is the default mode for hardware channels which lack support for digital formats.
    • passthrough — Channel mixing is done using fixed format/rate, but advanced operations such as digital passthrough also work. All channels will produce sound as usual until a digital format playback is requested. When this happens all other channels will be muted and the latest incoming digital format will be allowed to pass through undisturbed. Multiple concurrent digital streams are supported, but the latest stream will take precedence and mute all other streams.
    • adaptive — Works like the “passthrough” mode, but is a bit smarter, especially for multiple sound channels with different format/rate. When a new channel is about to start, the entire list of virtual channels will be scanned, and the channel with the best format/rate (usually the highest/biggest) will be selected. This ensures that mixing quality depends on the best channel. The downside is that the hardware DMA mode needs to be restarted, which may cause annoying pops or clicks.
  • dev.pcm.%d.[play|rec].vchanrate — sample rate speed for VCHAN mixing. All playback paths will be converted to this sample rate before the mixing process begins. 🎶
  • dev.pcm.%d.polling — experimental polling mode support where the driver operates by querying the device state on each tick using a callout(9) mechanism. Disabled by default and currently only available for a few device drivers. 🧚

🎙️ Recording Channels 🎙️

On devices that have more than one recording source (ie: mic and line), there is a corresponding /dev/dsp%d.r%d device. The mixer(8) utility can be used to start and stop recording from an specific device. 💖

📊 Statistics 📊

Channel statistics are only kept while the device is open. So with situations involving overruns and underruns, consider the output while the errant application is open and running. periodt 💅

🎛️ IOCTL Support 🎛️

The driver supports most of the OSS ioctl() functions, and most applications work unmodified. A few differences exist, while memory mapped playback is supported natively and in Linux emulation, memory mapped recording is not due to VM system design. As a consequence, some applications may need to be recompiled with a slightly modified audio module. See <sys/soundcard.h> for a complete list of the supported ioctl() functions. 🔥

📁 FILES 📁

The sound drivers may create the following device nodes:

  • /dev/audio%d.%d — Sparc-compatible audio device.
  • /dev/dsp%d.%d — Digitized voice device.
  • /dev/dspW%d.%d — Like /dev/dsp, but 16 bits per sample.
  • /dev/dsp%d.p%d — Playback channel.
  • /dev/dsp%d.r%d — Record channel.
  • /dev/dsp%d.vp%d — Virtual playback channel.
  • /dev/dsp%d.vr%d — Virtual recording channel.
  • /dev/sndstat — Current sound status, including all channels and drivers.

The first number in the device node represents the unit number of the sound device. All sound devices are listed in /dev/sndstat. Additional messages are sometimes recorded when the device is probed and attached, these messages can be viewed with the dmesg(8) utility. 🦋

The above device nodes are only created on demand through the dynamic devfs(5) clone handler. Users are strongly discouraged to access them directly. For specific sound card access, please instead use /dev/dsp or /dev/dsp%d. 🙈

✨ EXAMPLES ✨

Use the sound metadriver to load all sound bridge device drivers at once (for example if it is unclear which the correct driver to use is):

kldload snd_driver

Load a specific bridge device driver, in this case the Intel High Definition Audio driver:

kldload snd_hda

Check the status of all detected sound devices:

cat /dev/sndstat

Change the default sound device, in this case to the second device. This is handy if there are multiple sound devices available:

sysctl hw.snd.default_unit=1

🩺 DIAGNOSTICS 🩺

  • pcm%d:play:%d:dsp%d.p%d: play interrupt timeout, channel dead — The hardware does not generate interrupts to serve incoming (play) or outgoing (record) data.
  • unsupported subdevice XX — A device node is not created properly.

👀 SEE ALSO 👀

snd_ad1816(4), snd_ai2s(4), snd_als4000(4), snd_atiixp(4), snd_audiocs(4), snd_cmi(4), snd_cs4281(4), snd_csa(4), snd_davbus(4), snd_ds1(4), snd_emu10k1(4), snd_emu10kx(4), snd_envy24(4), snd_envy24ht(4), snd_es137x(4), snd_ess(4), snd_fm801(4), snd_gusc(4), snd_hda(4), snd_hdspe(4), snd_ich(4), snd_maestro(4), snd_maestro3(4), snd_mss(4), snd_neomagic(4), snd_sbc(4), snd_solo(4), snd_spicds(4), snd_t4dwave(4), snd_uaudio(4), snd_via8233(4), snd_via82c686(4), snd_vibes(4), devfs(5), device.hints(5), loader.conf(5), dmesg(8), kldload(8), mixer(8), sysctl(8)

📜 HISTORY 📜

The sound device driver first appeared in FreeBSD 2.2.6 as pcm, written by Luigi Rizzo. It was later rewritten in FreeBSD 4.0 by Cameron Grant. The API evolved from the VOXWARE standard which later became OSS standard. 🎀

👑 AUTHORS (the girlies behind the magic) 💖

Okay so this page has lore, bestie, and the lore is stacked:

  • Luigi Rizzo <luigi@iet.unipi.it> — he initially wrote the pcm device driver and this manual page. Founding mother energy, honestly. 🫶
  • Cameron Grant <gandalf@vilnya.demon.co.uk> — he later revised the device driver for FreeBSD 4.0. Glow-up arc fr fr. ✨
  • Seigo Tanimura <tanimura@r.dl.itc.u-tokyo.ac.jp> — revised this manual page. Editor icon. 💅

And then… it was rewritten for FreeBSD 5.2. Not me romanticising a version number but here we are. 😭

💀 BUGS (the not-so-slay part)

Some features of your sound card — like, for example, global volume control — might not be supported on all devices.

So like… vibe check your hardware before you come for the driver, okay? It's giving "it's not a bug, it's your sound card." 🙈 No cap, that's the whole warning. Lowkey heartbreaking, highkey important.

and that's the whole page, bestie!! 🎀✨

This one's from Debian, dated March 22, 2012, and it lives under the name SOUND(4) — section four, where the drivers hang out. 🦋 Go read it again if you need to, I'll wait. 💜

📦 where this page comes from