FILES.CONF(5)                 File Formats Manual                FILES.CONF(5)


     files.conf - rules base for the config utility


     The various files.* files located in the kernel source tree contain all
     the necessary information needed by config(8) to parse a kernel
     configuration file and determine the list of files to compile.

     The files.* rules base are simple, human-readable, text files.  Empty
     lines, as well as text prefixed by the `#' character, are ignored.


     The OpenBSD kernel "sees" the various devices as a hierarchical tree,
     where the various devices "attach" to parent entities, which can either
     be physical devices themselves (such as a computer bus), or logical
     entities, designed to make the driver code simpler.  Usually, the top-
     most devices are attached to the pseudo-device "mainbus", which is itself
     reported as attached to a fictitious "root" node.  There is no
     restriction on the "children" a device node may have; some device drivers
     can attach themselves to different kinds of parent devices.  For example,
     the logical scsibus(4) device can either attach at a SCSI controller
     device, or at the logical atapiscsi(4) bus.


     Some device attachments need to provide attachment information.  For
     example, an isa(4) device will use a range of I/O ports, one or more DMA
     channels, and one interrupt vector.  This attachment information is known
     as the "locators" for the device.  Most of the buses support default
     values for unspecified locators, for devices that either do not require
     them (such as isa(4) cards not using interrupts), or which can
     autoconfigure themselves (such as pci(4) devices).

     Attachment lines in the kernel configuration file must match the locators
     of the device they are attaching to.  For example:

           define pci {[dev = -1], [function = -1]}

     With the rule above, the following kernel configuration lines are valid:

           pciknob0 at pci? dev 2 function 42 # use fixed values
           pciknob* at pci? dev ? function ?  # use default values
           pciknob* at pci?                   # use default locators

     But the following are not:

           pciknob* at pci? trick ? treat ?               # unknown locators
           pciknob* at pci? dev ? function ? usefulness ? # unknown locators


     The following syntax defines a simple attribute, which can be later used
     to factorize code dependencies:

           define attribute

     An attachment-like attribute will also require locators to be specified.
     If no locators are necessary:

           define attribute {}

     If locators are provided:

           define attribute {[locator1 = default1], [locator2 = default2]}


     For simple device attachment, the following syntax defines a simple
     device, with no locators:

           device devicename {}

     If locators are necessary, they are specified as:

           device devicename {[locator1 = default1], [locator2 = default2]}

     A device can also reference an attribute with locators.  This is in fact
     a dependency rule.  For example, sys/conf/files defines the following
     attribute for SCSI controllers:

           define scsi {}           # no locators

     And SCSI drivers can then be defined as:

           device scsictrl: scsi

     A device may depend on as many attributes as necessary:

           device complexdev: simpledev, otherdev, specialattribute


     Pseudo devices are defined as regular devices, except that they do not
     need locators, and use a different keyword:

           pseudo-device loop: inet
           pseudo-device ksyms

     The rules above define, respectively, the loopback network interface and
     the kernel symbols pseudo-device.


     Due to the tree structure of the device nodes, every device but the
     pseudo devices need to attach to some parent node.  A device driver has
     to specify to which parents it can attach, with the following syntax:

           attach device at parent, parent2, parent3

     The rule above lists all the parent attributes a device may attach to.
     For example, given the following:

           device smartknob: bells, whistles
           attach smartknob at brainbus

     The following configuration line is then valid:

           smartknob* at brainbus?

     Whilst the following is not:

           smartknob* at dumbbus?

     If a device supports attachments to multiple parents, using different
     "glue" routines every time, the following syntax specifies the details:

           attach device at parent with device_parent_glue
           attach device at parent2 with device_parent2_glue

     This will define more required attributes, depending on the kernel
     configuration file's contents.


     It is possible to include other rules files anywhere in a file, using the
     "include" keyword:

           include "dev/pci/files.pci"

     The above rule will include the rules for machine-independent PCI code.

     The files sys/arch/machine/conf/files.machine, for every "machine" listed
     in the machine line in the kernel configuration file, as well as
     sys/conf/files, are always processed, and do not need to be included.


     The kernel configuration file description passed to config(8) lists
     several compilation options, as well as several device definitions.  From
     this list, config(8) will build a list of required attributes, which are:

           o   The "option" lines, with the option name translated to
               lowercase (for example, an "option INET6" line will produce the
               "inet6" attribute).
           o   The device and pseudo-device names, except for "root".


     Kernel source files are defined as:

           file file-list  dependencies    need-rules

     "file-list" typically only specifies a single filename.  If instead it
     contains a list of filenames separated by the `|' character, config(8)
     will select the first file from the list which exists.  If
     "${MACHINE_ARCH}" or "${MACHINE}" is found in the filename, it will be
     substituted with the relevant base architecture name.

     If the "dependencies" part is empty, the file will always be compiled in.
     This is the case for the core kernel files.  Otherwise, the file will
     only be added to the list if the dependencies are met.  Dependencies are
     based upon attributes and device names.  Multiple dependencies can be
     written using the "|" and "&" operators.  For example:

           file netinet/ipsec_input.c    (inet | inet6) & ipsec

     The above rule teaches config(8) to only add sys/netinet/ipsec_input.c to
     the filelist if the "ipsec" attribute, and at least one of the "inet" and
     "inet6" attributes, are required.

     The "need" rules can be empty, or one of the following keywords:

     needs-flag   Create an attribute header file, defining whether or not
                  this attribute is compiled in.
     needs-count  Create an attribute header file, defining how many instances
                  of this attribute are to be compiled in.  This rule is
                  mostly used for pseudo-devices.

     The "attribute header files" are simple C header files created in the
     kernel compilation directory, with the name attribute.h and containing
     the following line:

           #define NATTRIBUTE  0

     This would substitute the attribute name and its uppercase form, prefixed
     with the letter "N", to "attribute" and "NATTRIBUTE", respectively.  For
     a "needs-flag" rule, the value on the "#define" line is either 1 if the
     attribute is required, or 0 if it is not required.  For a "needs-count"
     rule, the value is the number of device instances required, or 0 if the
     device is not required.

     Attribute files are created for every attribute listed with a "need"
     rule, even if it is never referenced from the kernel configuration file.


     sys/arch/machine/conf/files.machine must also supply the following
     special commands:

     maxpartitions  Defines how many partitions are available on disk block
                    devices, usually 16.  This value is used by config(8) to
                    set up various device information structures.

     maxusers       Defines the bounds, and the default value, for the
                    "maxusers" parameter in the kernel configuration file.
                    The usual values are 2 8 64 ; config(8) will report an
                    error if the "maxusers parameter" in the kernel
                    configuration file does not fit in the specified range.


             Rules for architecture-dependent files, for the "machine"

             Rules for the "class" class of devices.

             Rules for the common SCSI subsystem.



OpenBSD 6.2                      March 2, 2016                     OpenBSD 6.2

[Unix Hosting | Open-Source | Contact Us]
[Engineering & Automation | Software Development | Server Applications]