gcc and friends. Generally you will need to install the package gcc. There may be cases where you will need a specific version of gcc to build kernel modules.

The following may be useful when testing the package or when preparing a package for a binary distribution (such as an rpm package) installing onto a subtree rather than on the real system.

This can be useful for any partial install target of the above (e.g: install-modules or install-programs).

the targetdir must be an absolute path, at least if you install the modules. To install to a relative path you can use something like:

To build extra modules / modules directory not included in the DAHDI distribution, use the optional variables MODULES_EXTRA and SUBDIRS_EXTRA:

Userspace programs communicate with the DAHDI modules through special device files under /dev/dahdi . Those are normally created by udev, but if you use an embedded-type system and don’t wish to use udev, you can generate them with the following script, from the source directory:

This will generate the device files under /dev/dahdi . If you need to generate them elsewhere (e.g. tmp/newroot/dev/dahdi) use the option -d to the script:

DKMS, Dynamic Kernel Module Support, is a framework for building Linux kernel modules. It is used, among others, by several distributions that package the DAHDI kernel modules.

DKMS is designed to provide updates over drivers installed from original kernel modules tree. Thus it installed modules into /lib/modules/updates or /lib/modules/VERSION/updates . This is generally not an issue on normal operation. However if you try to install DAHDI from source on a system with DAHDI installed from DKMS this way (potentially of an older version), be sure to remove the DKMS-installed modules from the updates directory. If you’re not sure, the following command will give you a clue of the versions installed:

DAHDI includes the wcb4xxp driver for the B410P, however, support for the B410P was historically provided by mISDN. If you would like to use the mISDN driver with the B410P, please comment out the wcb4xxp line in /etc/dahdi/modules. This will prevent DAHDI from loading wcb4xxp which will conflict with the mISDN driver.

To install the mISDN driver for the B410P, please see http://www.misdn.org for more information, but the following sequence of steps is roughly equivalent to make b410p from previous releases.

You will then also want to make sure /etc/init.d/misdn-init is started automatically with either chkconfig --add misdn-init or update-rc.d misdn-init defaults 15 30 depending on your distribution.

OSLEC is an Open Source Line Echo Canceller. It is currently in the staging subtree of the mainline kernel and will hopefully be fully merged at around version 2.6.29. The echo canceller module dahdi_echocan_oslec provides a DAHDI echo canceller module that uses the code from OSLEC. As OSLEC has not been accepted into mainline yet, its interface is not set in stone and thus this driver may need to change. Thus it is not built by default.

Luckily the structure of the dahdi-linux tree matches that of the kernel tree. Hence you can basically copy drivers/staging/echo and place it under driver/staging/echo . In fact, dahdi_echocan_oslec assumes that this is where the oslec code lies. If it is elsewhere you’ll need to fix the #include line.

Thus for the moment, the simplest way to build OSLEC with dahdi is to copy the directory drivers/staging/echo from a recent kernel tree (at least 2.6.28-rc1) to the a subdirectory with the same name in the dahdi-linux tree.

After doing that, you’ll see the following when building (running make)

As this is an experimental driver, problems building and using it should be reported on the OSLEC mailing list.

Alternatively you can also get the OSLEC code from the dahdi-linux-extra GIT repository:

In many cases you already have DAHDI installed on your system but would like to try a different version. E.g. in order to check if the latest version fixes a bug that your current system happens to have.

DAHDI-linux includes a script to automate the task of installing DAHDI to a subtree and using it instead of the system copy. Module loading through modprobe cannot be used. Thus the script pre-loads the required modules with insmod (which requires some quesswork as for which modules to load). It also sets PATH and other environment variables to make all the commands do the right thing.

There is an extra mode of operation to copy all the required files to a remote host and run things there, for those who don’t like to test code on thir build system.

(most modules)

Sets debug mode / debug level. With most modules debug can be either disabled (0, the default value) or enabled (any other value).

wctdm and wcte1xp print several extra debugging messages if the value of debug is more than 1.

Some modules have "debugging flags" bits - the value of debug is a bitmask and several messages are printed if some bits are set:

To get a list of parameters supported by a module, use

Or, for a module you have just built:

For the xpp modules this will also include the description and default value of the module. You can find a list of useful xpp module parameters in README.Astribank .

(dahdi)

The default size for the echo canceller. The number is in "taps", that is "samples", 1/8 ms. The default is 64 - for a tail size of 8 ms.

Asterisk’s chan_dahdi tends to pass its own value anyway, with a different default size. So normally setting this doesn’t change anything.

(dahdi)

The maximum number of pseudo channels that dahdi will allow userspace to create. Pseudo channels are used when conferencing channels together. The default is 512.

(dahdi)

See Span Assignments below.

Userspace programs will usually interact with DAHDI through device files under the /dev/dahdi directory (pedantically: character device files with major number 196) . Those device files can be generated statically or dynamically through the udev system.

A PBX system should generally have a single clock. If you are connected to a telephony provider via a digital interface (e.g: E1, T1) you should also typically use the provider’s clock (as you get through the interface). Hence one important job of Asterisk is to provide timing to the PBX.

DAHDI "ticks" once per millisecond (1000 times per second). On each tick every active DAHDI channel reads and 8 bytes of data. Asterisk also uses this for timing, through a DAHDI pseudo channel it opens.

However, not all PBX systems are connected to a telephony provider via a T1 or similar connection. With an analog connection you are not synced to the other party. And some systems don’t have DAHDI hardware at all. Even a digital card may be used for other uses or is simply not connected to a provider. DAHDI cards are also capable of providing timing from a clock on card. Cheap x100P clone cards are sometimes used for that purpose.

If a hardware timing source either cannot be found or stops providing timing during runtime, DAHDI will automatically use the host timer in order provide timing.

You can check the DAHDI timing source with dahdi_test, which is a small utility that is included with DAHDI. It runs in cycles. In each such cycle it tries to read 8192 bytes, and sees how long it takes. If DAHDI is not loaded or you don’t have the device files, it will fail immediately. If you lack a timing device it will hang forever in the first cycle. Otherwise it will just give you in each cycle the percent of how close it was. Also try running it with the option -v for a verbose output.

DAHDI provides telephony channels to the userspace applications. Those channels are channels are incorporated into logical units called spans.

With digital telephony adapters (e.g: E1 or T1), a span normally represents a single port. With analog telephony a span typically represents a PCI adapter or a similar logical unit.

Both channels and spans are identified by enumerating numbers (beginning with 1). The number of the channel is the lowest unused one when it is generated, and ditto for spans.

There are up to 128 spans and 1024 channels. This is a hard-wired limit (see dahdi/user.h . Several places throuout the code assume a channel number fits in a 16 bits number). Channel and span numbers start at 1.

A DAHDI device (e.g. a PCI card) is represented within the DAHDI drivers as a DAHDI device. Normally (with auto_assign_spans=1 in the module dahdi, which is the default), when a device is discovered and loaded, it registers with the DAHDI core and its spans automatically become available. However if you have more than one device, you may be interested to set explicit spans and channels numbers for them. To use manual span assigment, set auto_assign_spans to 0 . e.g. in a file under /etc/modprobe.d/ include the following line:

You will then need to assign the spans manually at device startup. You will need to assign a span number and channel numbers for each available span on the system. On my test system I have one BRI PCI card and one Astribank BRI+FXS:

All spans here, except the FXS one, are BRI spans with 3 channels per span.

In order to assign a span, we write three numbers separated by colns to the file assign_span in the SysFS node

Thus:

As you can see, the order of span numbers or local span number is insignificant. However the order of channel numbers must be the same as that of span numbers.

Which indeed produced:

Likewise spans can be unassigned by writing to the unassign-span "file".

Dynamic spans are spans that are not represented by real hardware. Currently there are two types of them:

Modules that support the dynamic spans are typically loaded at the time the ioctl DAHDI_DYNAMIC_CREATE is called. This is typically called by dahdi_cfg when it has a line such as:

in /etc/dahdi/system.conf . In that case it will typically try to load (through modprobe) the modules dahdi_dynamic and dahdi_dynamic_’somename'. It will then pass params to it.

Dynamic spans are known to be tricky and are some of the least-tested parts of DAHDI.

(To be documented later)

(To be documented later)

A simple way to get the current list of spans and channels each span contains is the files under /proc/dahdi . /proc/dahdi is generated by DAHDI as it loads. As each span registers to DAHDI, a file under /proc/dahdi is created for it. The name of that file is the number of that span.

Each file has a 1-line title for the span followed by a few optional general counter lines, an empty line and then a line for each channel of the span.

The title line shows the number of the span, its name and title, and (potentially) the alarms in which it is.

The title shows the span number and name, followed by any alarms the span may have: For example, here is the first span in my system (with no alarms):

There are several extra optional keywords that may be added there:

Following that line there may be some optional lines about IRQ misses, timing slips and such, if there are any.

The channel line for each channel shows its channel number, name and the actual signalling assigned to it through dahdi_cfg. Before being configured by dahdi_cfg: This is DAHDI channel 2, whose name is XPP_FXS/0/0/1.

After being configured by dahdi_cfg: the signalling FXOLS was added. FXS channels have FXO signalling and vice versa:

If the channel is in use (typically opened by Asterisk) then you will see an extra (In use):

DAHDI exposes several interfaces under the SysFS virtual file system. SysFS represents kernel objects in nodes - directories. There properties are often files. They may also contain other nodes or include symlinks to other nodes.

User-space programs can only work with DAHDI channels. The basic operation is read() to read audio from the device and write() to write audio to it. Audio can be encoded as either alaw (G.711a) or (m)ulaw (G.711u). See the ioctl DAHDI_SETLAW.

While it is technically possible to use /dev/dahdi/NUMBER directly, this will only work for channel numbers up to 249. Generally you should use:

FIXME: there’s more to tell about the user-space interface.

Most of the configuration is applied from userspace using the tool dahdi_cfg in the package dahdi_tools. This section will not cover the specifics of that file. Rather it will cover the way configuration from this file is applied. Also note that there are other methods to configure DAHDI drivers: there are Module Parameters. The xpp driver have their own separate initialization scripts and xpp.conf that arecovered in README.Astribank.

When a span is registered, it is considered "unconfigured". Only after dahdi_cfg has been run to apply configuration, the span is ready to run.

Some of the configuration is handled by the DAHDI core. Some of it is handled by callbacks, which are function pointers in the ‘struct dahdi_span’: spanconfig, chanconfig and (in a way) startup.

Dahdi_cfg starts by reading the configuration from the configuration file (/etc/dahdi/system.conf, by default), and creating a local configuration to apply. If you use -v, at this stage it will pront the configuration that is "about to be configured". Then it will start to actually apply the configuration.

Before actually applying configuration, it destroys any existing dynamic spans and generates new ones (if so configured. FIXME: what about running DAHDI_SPANCONFIG for new dynamic spans?).

Next thing it will do is apply the parameters from span lines using the ioctl DAHDI_SPANCONFIG. Some generic processing of parameters passed in DAHDI_SPANCONFIG is done in the DAHDI core, in the callback function spanconfig in , but most of it is left to spanconfig callback of the span (if it exists. This one typically does not exists on analog cards).

Now dahdi_cfg will ask each existing channel for its existing configuration and apply configuration if configuration changes are needed. Configuration is applied to the channels using the ioctl call DAHDI_CHANCONFIG ioctl. As in the case of the span configuration, part of it is applied by the DAHDI core, and then it is handed over to the span’s chanconfig callback. Typically all spans will have such a callback.

Echo cancellers and tone-zones are handled separately later.

Once everything is done, the ioctl DAHDI_STARTUP is called for every span. This is also translated to a call to the optional span callback startup.

Finally the ioctl DAHDI_HDLC_RATE is called for every channel (that is: if 56k is not set for the channel, it will be explicitly set to the standard HDLC rate of 64k).

Low-level drivers create spans (struct dahdi_span). They register the spans with the DAHDI core using dahdi_device_register().

struct dahdi_span has a number of informative members that are updated solely by the low-level driver:

There are various function pointers in the struct dahdi_span which are used by the DAHDI core to call the low-level driver. Most of them are optional.

The following apply to a span:

The following apply to a single channel.

There are several alternative methods for a span to use for signalling. One of them should be used.

Like any other kernel code, DAHDI strives to maintain a stable interface to userspace programs. The API of DAHDI to userspace programs, dahdi/user.h, has remained backward-compatible for a long time and is expected to remain so in the future. With the ABI (the bits themselves) things are slightly trickier.

DAHDI’s interface to userspace is mostly ioctl(3) calls. Ioctl calls are identified by a number that stems from various things, one of which is the size of the data structure passed between the kernel and userspace.

Many of the DAHDI ioctl-s use some specific structs to pass information between kernel and userspace. In some cases the need arose to pass a few more data members in each call. Simply adding a new member to the struct would have meant a new number for the ioctl, as its number depends on the size of the data passed.

Thus we would add a new ioctl with the same base number and with the original struct.

So suppose we had the following ioctl:

And we want to add the field int onemore, we won’t just add it to the struct. We will do something that is more complex:

We actually have here two different ioctls: the old DAHDI_EXAMPLE would be 0xC004DA3E . DAHDI_EXAMPLE_V1 would have the same value. But the new value of DAHDI_EXAMPLE would be 0xC008DA3E .

Programs built with the original dahdi/user.h (before the change) use the original ioctl, whether or not the kernel code is actually of the newer version. Thus in most cases there are no compatibility issues.

When can we have compatibility issues? If we have code built with the new dahdi/user.h, but the loaded kernel code (modules) are of the older version. Thus the userspace program will try to use the newer DAHDI_EXAMPLE (0xC008DA3E). But the kernel code has no handler for that ioctl. The result: the error 25, ENOTTY, which means "Inappropriate ioctl for device".

As a by-product of that method, for each interface change a new #define is added. That definition is for the old version and thus it might appear slightly confusing in the code, but it is useful for writing code that works with all versions of DAHDI.

An alarm indicates that a port is not available for some reason. Thus it is probably not a good idea to try to call out through it.

KB1 was not designed to function at greater than 128 taps, and if configured this way, will result in the destruction of audio. Ideally DAHDI would return an error when a KB1 echocanceller is configured with greater than 128 taps.