path: root/en_US.ISO_8859-1/tutorials/ddwg/ddwg.sgml

<!DOCTYPE linuxdoc PUBLIC "-//FreeBSD//DTD linuxdoc//EN">

<!--
  ++++++++++++++++++++++++++++++++++++++++++++++++++
  ++ file: /home/erich/lib/src/sgml/ddwg.sgml
  ++
  ++ Copyright Eric L. Hernes - Wednesday, August 2, 1995
  ++
  ++ $FreeBSD$
  ++
  ++ Sgml doc for something
  -->

<article>

<title>FreeBSD Device Driver Writer's Guide
<author>Eric L. Hernes, <tt/erich@rrnet.com/
<date>Wednesday, May 29, 1996

<abstract>

This document describes how to add a device driver to FreeBSD.  It is
<it/not/ intended to be a tutorial on UNIX device drivers in general.
It is intended for device driver authors, familiar with the UNIX
device driver model, to work on FreeBSD.

</abstract>

<toc>

<sect>2.x specific

<p>Due to changes in FreeBSD over time, this guide is only accurate as
regards FreeBSD 2.x.  A replacement guide for FreeBSD 3.x and beyond
is being written.  Please contact Jeroen Ruigrok <tt/&lt;asmodai@wxs.nl&gt;/
if you would like to help with this.

<sect> Overview

<p> <it> 
The FreeBSD kernel is very well documented, unfortunately it's all
in `C'.
</it>

<sect> Types of drivers.

<sect1> Character

<sect2> Data Structures
<p> <tt/struct cdevsw/ Structure

<sect2> Entry Points
<sect3> d_open()
<p>
d_open() takes several arguments, the formal list looks something like:
<code>
int
d_open(dev_t dev, int flag, int mode, struct proc *p)
</code>
d_open() is called on <em/every/ open of the device.
<p>

The <tt/dev/ argument contains the major and minor number of the
device opened.  These are available through the macros <tt/major()/ and
<tt/minor()/
<p>

The <tt/flag/ and <tt/mode/ arguments are as described in the 
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?open(2)" name="open(2)">
manual page.  It is recommended that you check these for access modes
in &lt;sys/fcntl.h&gt; and do what is required.  For example if <tt/flag/
is (O_NONBLOCK | O_EXLOCK) the open should fail if either it would
block, or exclusive access cannot be granted.
<p>

The <tt/p/ argument contains all the information about the current
process.

<sect3> d_close()
<p>
d_close() takes the same argument list as d_open():
<code>
int
d_close(dev_t dev , int flag , int mode , struct proc *p)
</code>

d_close() is only called on the last close of your device (per minor
device). For example in the following code fragment, d_open() is called
3 times, but d_close() is called only once.
<code>
 ...
    fd1=open("/dev/mydev", O_RDONLY);
    fd2=open("/dev/mydev", O_RDONLY);
    fd3=open("/dev/mydev", O_RDONLY);
 ...
   <useful stuff with fd1, fd2, fd3 here>
 ...
    close(fd1);
    close(fd2);
    close(fd3);
 ...
</code>

The arguments are similar to those described above for 
d_open().

<sect3> d_read() and d_write()
<p>
d_read() and d_write take the following argument lists:
<code>
int
d_read(dev_t dev, struct uio *uio, int flat)
int
d_write(dev_t dev, struct uio *uio, int flat)
</code>

The d_read() and d_write() entry points are called when
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?read(2)" name="read(2)"> and
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?write(2)" name="write(2)">
are called on your device from user-space.  The transfer
of data can be handled through the kernel support routine uiomove().

<sect3> d_ioctl()
<p>
It's argument list is as follows:
<code>
int
d_ioctl(dev_t dev, int cmd, caddr_t arg, int flag, struct proc *p)
</code>

d_ioctl() is a catch-all for operations which don't make sense in
a read/write paradigm.  Probably the most famous of all ioctl's is on
tty devices, through
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?stty(1)" name="stty(1)">.
The ioctl entry point is called from
ioctl() in sys/kern/sys_generic.c<p>

There are four different types of ioctl's which can be implemented.
&lt;sys/ioccom.h&gt; contains convenience macros for defining these ioctls.

<tt/_IO(g,n)/ for control type operations. &nl;

<tt/_IOR(g,n,t)/ for operations that read data from a device. &nl;

<tt/_IOW(g,n,t)/ for operations that write data to a device. &nl;

<tt/_IOWR(g,n,t)/ for operations that write to a device, and then 
read data back. &nl;

Here <tt/g/ refers to a <em/group/. This is an 8-bit value, typically
indicative of the device; for example, 't' is used in tty ioctls.
<tt/n/ refers to the number of the ioctl within the group.  On SCO, this
number alone denotes the ioctl.  <tt/t/ is the data type which will
get passed to the driver; this gets handed to a sizeof() operator in
the kernel.  The ioctl() system call will either copyin() or copyout()
or both for your driver, then hand you a pointer to the data structure
in the <tt/arg/ argument of the d_ioctl call.  Currently the data size
is limited to one page (4k on the i386).

<sect3> d_stop()
<sect3> d_reset()
<sect3> d_devtotty()
<sect3> d_poll() (3.0+) or d_select() (2.2)
<p>
d_poll()'s argument list is as follows:
<code>
void
d_poll(dev_t dev, int events, struct proc *p)
</code>

<p> d_poll() is used to find out if a device is ready for IO.  An
example is waiting for data to become available from the network, or
waiting for the user to press a key.  This correspond to the poll()
call in userland.

<p>The d_poll() call should check for the events specified in the
event mask.  If none of the requested events are active, but they may
become active later, it should record this for the kernel's later
persual.  d_poll() does this by calling selrecord() with a selinfo
structure for this device.  The sum of all these activities look
something like this:

<code>
static struct my_softc {
	struct queue rx_queue;  /* As example only - not required */
	struct queue tx_queue;  /* As example only - not required */
	struct selinfo selp;    /* Required */
} my_softc[NMYDEV];

...

static int
mydevpoll(dev_t dev, int events, struct proc *p)
{
	int revents = 0;	/* Events we found */
	int s;
	struct my_softc *sc = &amp;my_softc[dev];

	/* We can only check for IN and OUT */
	if ((events &amp; (POLLIN|POLLOUT)) == 0)
		return(POLLNVAL);
	
	s = splhigh();
	/* Writes are if the transmit queue can take them */
	if ((events &amp; POLLOUT) &amp;&amp;
	    !IF_QFULL(sc->tx_queue))
		revents |= POLLOUT;
	/* ... while reads are OK if we have any data */
	if ((events &amp; POLLIN) &amp;&amp;
	    !IF_QEMPTY(sc->rx_queue))
		revents |= POLLIN;
	if (revents == 0)
		selrecord(p, &amp;sc->selp);
	splx(s);
	return revents;
}
</code>

<p>
d_select() is used in 2.2 and previous versions of FreeBSD.  Instead
of 'events' it take a single int 'rw', which can be FREAD for reads
(as in POLLIN above), FWRITE for write (as in POLLOUT above), and 0
for 'exception' - something exceptional happened, like a card being
inserted or removed for the pccard driver.

<p>For select, the above code fragment would look like this:
<code>
static int
mydevselect(dev_t dev, int rw, struct proc *p)
{
	int ret = 0;
	int s;
	struct my_softc *sc = &amp;my_softc[dev];

	s = splhigh();
	switch (rw) {
	case FWRITE:
		/* Writes are if the transmit queue can take them */
		if (!IF_QFULL(sc->tx_queue))
			ret = 1;
		break;
	case FREAD:
		/* ... while reads are OK if we have any data */
		if (!IF_QEMPTY(sc->rx_queue))
			ret = 1;
		break;
	case 0:
		/* This driver never get any exceptions */
		break;
        }
	if(ret == 0)
		selrecord(p, &amp;sc->selp);
	splx(s);
	return(revents);
}
</code>

<sect3> d_mmap()
<sect3> d_strategy()
<p>
d_strategy()'s argument list is as follows:
<code>
void
d_strategy(struct buf *bp)
</code>

<p> d_strategy() is used for devices which use some form of scatter-gather
io.  It is most common in a block device.  This is significantly different
than the System V model, where only the block driver performs scatter-gather
io. Under BSD, character devices are sometimes requested to perform
scatter-gather io via the readv() and writev() system calls.

<sect2> Header Files

<sect1> Block
<sect2> Data Structures
<p>  <tt/struct bdevsw/ Structure
<p>  <tt/struct buf/ Structure

<sect2> Entry Points
<sect3> d_open()
<p> Described in the Character device section.

<sect3> d_close()
<p> Described in the Character device section.

<sect3> d_strategy()
<p> Described in the Character device section.

<sect3> d_ioctl()
<p> Described in the Character device section.

<sect3> d_dump()

<sect3> d_psize()

<sect2> Header Files

<sect1> Network
<sect2> Data Structures
<p> <tt/struct ifnet/  Structure

<sect2> Entry Points
<sect3> if_init()
<sect3> if_output()
<sect3> if_start()
<sect3> if_done()
<sect3> if_ioctl()
<sect3> if_watchdog()

<sect2> Header Files

<sect1> Line Discipline
<sect2> Data Structures

<p> <tt/struct linesw/ Structure

<sect2> Entry Points
<sect3> l_open()
<sect3> l_close()
<sect3> l_read()
<sect3> l_write()
<sect3> l_ioctl()
<sect3> l_rint()
<sect3> l_start()
<sect3> l_modem()

<sect2> Header Files

<sect> Supported Busses

<sect1> ISA -- Industry Standard Architecture
<sect2> Data Structures

<sect3> <tt/struct isa_device/ Structure
<p>
This structure is required, but generally it is created by
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)">
from the kernel configuration file.  It is required on a per-device
basis, meaning that if you have a driver which controls two serial
boards, you will have two isa_device structures.  If you build a
device as an LKM, you must create your own isa_device structure to
reflect your configuration. (lines 85 - 131 in pcaudio_lkm.c) There is
nearly a direct mapping between the config file and the isa_device
structure. The definition from /usr/src/sys/i386/isa/isa_device.h is:
<code>
struct isa_device {
        int     id_id;          /* device id */
        struct  isa_driver *id_driver;
        int     id_iobase;      /* base i/o address */
        u_short id_irq;         /* interrupt request */
        short   id_drq;         /* DMA request */
        caddr_t id_maddr;       /* physical i/o memory address on bus (if any)*/
        int     id_msize;       /* size of i/o memory */
        inthand2_t *id_intr;    /* interrupt interface routine */
        int     id_unit;        /* unit number */
        int     id_flags;       /* flags */
        int     id_scsiid;      /* scsi id if needed */
        int     id_alive;       /* device is present */
#define RI_FAST         1               /* fast interrupt handler */
        u_int   id_ri_flags;    /* flags for register_intr() */
        int     id_reconfig;    /* hot eject device support (such as PCMCIA) */
        int     id_enabled;     /* is device enabled */
        int     id_conflicts;   /* we're allowed to conflict with things */
        struct isa_device *id_next; /* used in isa_devlist in userconfig() */
};
</code>

<!-- XXX add stuff here -->
<sect3> <tt/struct isa_driver/ Structure

<p>
This structure is defined in ``/usr/src/sys/i386/isa/isa_device.h''.
These are required on a per-driver basis.  The definition is:
<code>
struct isa_driver {
        int     (*probe) __P((struct isa_device *idp));
                                        /* test whether device is present */
        int     (*attach) __P((struct isa_device *idp));
                                        /* setup driver for a device */
        char    *name;                  /* device name */
        int     sensitive_hw;           /* true if other probes confuse us */
};
</code>

This is the structure used by the probe/attach code to detect and
initialize your device.  The <tt/probe/ member is a pointer to your
device probe function; the <tt/attach/ member is a pointer to your
attach function.  The <tt/name/ member is a character pointer to the
two or three letter name for your driver.  This is the name reported
during the probe/attach process (and probably also in 
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?lsdev(8)" name="lsdev(8)">). The
<tt/sensitive_hw/ member is a flag which helps the probe code
determine probing order.

A typical instantiation is:
<code>
struct  isa_driver      mcddriver = { mcd_probe, mcd_attach, "mcd" };
</code>

<sect2> Entry Points

<sect3> probe()
<p>
probe() takes a <tt/struct isa_device/ pointer as an argument and returns
an int.  The return value is ``zero'' or ``non-zero'' as to the absence
or presence of your device.  This entry point may (and probably should)
be declared as <tt/static/ because it is accessed via the <tt/probe/ member
of the <tt/struct isa_driver/ structure.  This function is intended
to detect the presence of your device only; it should not do any
configuration of the device itself.

<sect3> attach()
<p>
attach() also takes a <tt/struct isa_device/ pointer as an argument and
returns an int.  The return value is also ``zero'' or ``non-zero'' indicating
whether or not the attach was successful.  This function is intended
to do any special initialization of the device as well as confirm that
the device is usable.  It too should be declared <tt/static/ because 
it is accessed through the <tt/attach/ member of the <tt/isa_driver/ 
structure.

<sect2> Header Files

<sect1> EISA -- Extended Industry Standard Architecture

<sect2> Data Structures

<p> <tt/struct eisa_dev/ Structure
<p> <tt/struct isa_driver/ Structure

<sect2> Entry Points

<sect3> probe()
<p> Described in the ISA device section.

<sect3> attach()
<p> Described in the ISA device section.

<sect2> Header Files

<sect1> PCI -- Peripheral Computer Interconnect
<sect2> Data Structures

<p> <tt/struct pci_device/ Structure

      name:   The short device name.

      probe:  Checks if the driver can support a device
              with this type. The tag may be used to get
              more info with pci_read_conf(). See below.
              It returns a string with the device's name,
              or a NULL pointer, if the driver cannot
              support this device.

      attach: Allocate a control structure and prepare
              it. This function may use the PCI mapping
              functions. See below.
              (configuration id) or type.

      count:  A pointer to a unit counter.
              It's used by the PCI configurator to
              allocate unit numbers.

<sect2> Entry Points

<sect3> probe()

<sect3> attach()

<sect3> shutdown()

<sect2> Header Files

<sect1> SCSI -- Small Computer Systems Interface
<sect2> Data Structures

<p> <tt/struct scsi_adapter/ Structure
<p> <tt/struct scsi_device/ Structure
<p> <tt/struct scsi_ctlr_config/ Structure
<p> <tt/struct scsi_device_config/ Structure
<p> <tt/struct scsi_link/ Structure

<sect2> Entry Points
<sect3> attach()
<sect3> init()

<sect2> Header Files

<sect1> PCCARD (PCMCIA)
<sect2> Data Structures
<p> <tt/struct slot_cont/ Structure
<p> <tt/struct pccard_drv/ Structure
<p> <tt/struct pccard_dev/ Structure
<p> <tt/struct slot/ Structure

<sect2> Entry Points
<sect3> handler()
<sect3> unload()
<sect3> suspend()
<sect3> init()

<sect2> Header Files
           a. &lt;pccard/slot.h&gt;

<sect> Linking Into the Kernel.

<p>
In FreeBSD, support for the ISA and EISA busses is i386 specific.
While FreeBSD itself is presently available on the i386 platform,
some effort has been made to make the PCI, PCCARD, and SCSI code
portable.  The ISA and EISA specific code resides in
/usr/src/sys/i386/isa and /usr/src/sys/i386/eisa respectively.
The machine independent PCI, PCCARD, and SCSI code reside in
/usr/src/sys/{pci,pccard,scsi}. The i386 specific code for these
reside in /usr/src/sys/i386/{pci,pccard,scsi}.

<p>
In FreeBSD, a device driver can be either binary or source.  There is
no ``official'' place for binary drivers to reside.  BSD/OS uses
something like sys/i386/OBJ.  Since most drivers are distributed
in source, the following discussion refers to a source driver.
Binary only drivers are sometimes provided by hardware vendors
who wish to maintain the source as proprietary.

<p>
A typical driver has the source code in one c-file, say dev.c.  The
driver also can have some include files; devreg.h typically contains
public device register declarations, macros, and other driver
specific declarations.  Some drivers call this devvar.h instead.
Some drivers, such as the dgb (for the Digiboard PC/Xe),
require microcode to be loaded onto the board.  For the dgb driver
the microcode is compiled and dumped into a header file ala
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?file2c(1)" name="file2c(1)">.

<p>
If the driver has data structures and ioctl's which are specific to
the driver/device, and need to be accessible from user-space, they
should be put in a separate include file which will reside in
/usr/include/machine/ (some of these reside in /usr/include/sys/).
These are typically named something like ioctl_dev.h or devio.h.

<p>
If a driver is being written which, from user space is
identical to a device which already exists, care should be taken to
use the same ioctl interface and data structures.  For example, from
user space, a SCSI CDROM drive should be identical to an IDE cdrom
drive; or a serial line on an intelligent multiport card (Digiboard,
Cyclades, ...)  should be identical to the sio devices.  These devices
have a fairly well defined interface which should be used.

<p>
There are two methods for linking a driver into the kernel, static and 
the LKM model.  The first method is fairly standard across the
*BSD family.  The other method was originally developed by Sun
(I believe), and has been implemented into BSD using the Sun model.
I don't believe that the current implementation uses any Sun code.

<sect1> Standard Model

<p>
The steps required to add your driver to the standard FreeBSD kernel are
<itemize>
<item> Add to the driver list
<item> Add an entry to the &lsqb;bc&rsqb;devsw
<item> Add the driver entry to the kernel config file
<item> <htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)">, 
compile, and install the kernel
<item> make required nodes.
<item> reboot.
</itemize>

<sect2> Adding to the driver list.
<p>
The standard model for adding a device driver to the Berkeley kernel
is to add your driver to the list of known devices. This list is 
dependent on the CPU architecture.  If the device is not i386 specific
(PCCARD, PCI, SCSI), the file is in ``/usr/src/sys/conf/files''.
If the device is i386 specific, use ``/usr/src/sys/i386/conf/files.i386''.
A typical line looks like:
<tscreen><code>
i386/isa/joy.c                  optional        joy     device-driver
</code></tscreen>

The first field is the pathname of the driver module relative to
/usr/src/sys. For the case of a binary driver the path would be
something like ``i386/OBJ/joy.o''.

The second field tells 
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)">
that this is an optional driver. Some
devices are required for the kernel to even be built.

The third field is the name of the device.

The fourth field tells config that it's a device driver (as opposed to
just optional).  This causes config to create entries for the device
in some structures in /usr/src/sys/compile/KERNEL/ioconf.c.

It is also possible to create a file
``/usr/src/sys/i386/conf/files.KERNEL'' whose contents will override
the default files.i386, but only for the kernel ``KERNEL''.

<sect2>Make room in conf.c
<p>
Now you must edit ``/usr/src/sys/i386/i386/conf.c'' to make an entry
for your driver.  Somewhere near the top, you need to declare your
entry points. The entry for the joystick driver is:
<code>
#include "joy.h"
#if NJOY > 0
d_open_t        joyopen;
d_close_t       joyclose;
d_rdwr_t        joyread;
d_ioctl_t       joyioctl;
#else
#define joyopen         nxopen
#define joyclose        nxclose
#define joyread         nxread
#define joyioctl        nxioctl
#endif
</code>

This either defines your entry points, or null entry points which
will return ENXIO when called (the #else clause).

The include file ``joy.h'' is automatically generated by
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)"> when
the kernel build tree is created.  This usually has only one line like:
<code>
#define NJOY 1
</code>
or
<code>
#define NJOY 0
</code>
which defines the number of your devices in your kernel.

You must additionally add a slot to either cdevsw&lsqb;&rsqb, or to
bdevsw&lsqb;&rsqb, depending on whether it is a character device or
a block device, or both if it is a block device with a raw interface.
The entry for the joystick driver is:

<code>
/* open, close, read, write, ioctl, stop, reset, ttys, select, mmap, strat */
struct cdevsw   cdevsw[] =
{
 ...
        { joyopen,      joyclose,       joyread,        nowrite,        /*51*/
          joyioctl,     nostop,         nullreset,      nodevtotty,/*joystick */
          seltrue,      nommap,         NULL},
 ...
}
</code>

Order is what determines the major number of your device. Which is why
there will always be an entry for your driver, either null entry
points, or actual entry points.  It is probably worth noting that this
is significantly different from SCO and other system V derivatives,
where any device can (in theory) have any major number.  This is
largely a convenience on FreeBSD, due to the way device nodes are
created.  More on this later.

<sect2>Adding your device to the config file.
<p>
This is simply adding a line describing your device.
The joystick description line is:
<verb>
device          joy0    at isa? port "IO_GAME"
</verb>
This says we have a device called ``joy0'' on the isa bus using
io-port ``IO_GAME'' (IO_GAME is a macro defined in
/usr/src/sys/i386/isa/isa.h).

A slightly more complicated entry is for the ``ix'' driver:
<verb>
device ix0 at isa? port 0x300 net irq 10 iomem 0xd0000 iosiz 32768 vector ixintr
</verb>
This says that we have a device called `ix0' on the ISA bus.  It uses
io-port 0x300. It's interrupt will be masked with other devices in
the network class. It uses interrupt 10. It uses
32k of shared memory at physical address 0xd0000.  It also defines
it's interrupt handler to be ``ixintr()''

<sect2><htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)">
the kernel.
<p>
Now with our config file in hand, we can create a kernel compile directory.
This is done by simply typing:
<verb>
# config KERNEL
</verb>
where KERNEL is the name of your config file.  Config creates a
compile tree for you kernel in /usr/src/sys/compile/KERNEL.  It
creates the Makefile, some .c files, and some .h files with macros
defining the number of each device in your kernel.

Now you can go to the compile directory and build.  Each time you run
config, your previous build tree will be removed, unless you config
with a -n.  If you have config'ed and compiled a GENERIC kernel, you can
``make links'' to avoid compiling a few files on each iteration.  I typically
run
<verb>
# make depend links all
</verb>
followed by a ``make install'' when the kernel is done to my liking.

<sect2>Making device nodes.
<p>
On FreeBSD, you are responsible for making your own device nodes.  The
major number of your device is determined by the slot number in the
device switch.  Minor number is driver dependent, of course.  You can
either run the mknod's from the command line, or add a section to
/dev/MAKEDEV.local, or even /dev/MAKEDEV to do the work.  I sometimes
create a MAKEDEV.dev script that can be run stand-alone or pasted
into /dev/MAKEDEV.local

<sect2>Reboot.
<p>
This is the easy part.  There are a number of ways to do this, reboot,
fastboot, shutdown -r, cycle the power, etc. Upon bootup you should
see your XXprobe() called, and if all is successful, your XXattach()
too.

<sect1> Loadable Kernel Module (LKM)

<p>
There are really no defined procedures for writing an LKM driver.  The
following is my own conception after experimenting with the LKM device
interface and looking at the standard device driver model, this is one
way of adding an LKM interface to an existing driver without touching
the original driver source (or binary).  It is recommended though,
that if you plan to release source to your driver, the LKM specific
parts should be part of the driver itself, conditionally compiled
on the LKM macro (i.e. #ifdef LKM).

This section will focus on writing the LKM specific part of the driver.  We
will assume that we have written a driver which will drop into the standard
device driver model, which we would now like to implement as an LKM.  We will
use the pcaudio driver as a sample driver, and develop an LKM front-end.  The
source and makefile for the pcaudio LKM, ``pcaudio_lkm.c'' and ``Makefile'',
should be placed in /usr/src/lkm/pcaudio.  What follows is a breakdown of
pcaudio_lkm.c.

Lines  17 -  26  

  --  This includes the file ``pca.h'' and conditionally compiles the rest
of the LKM on whether or not we have a pcaudio device defined.  This
mimics the behavior of config.  In a standard device driver,
<htmlurl url="http://www.freebsd.org/cgi/man.cgi?config(8)" name="config(8)">
generates the pca.h file from the number pca devices in the config file.
<code>
    17  /* 
    18   * figure out how many devices we have..
    19   */
    20
    21  #include "pca.h"
    22
    23  /*
    24   * if we have at least one ...
    25   */
    26  #if NPCA > 0
</code>

Lines  27 -  37

 -- Includes required files from various include directories.
<code>
    27  #include <sys/param.h>
    28  #include <sys/systm.h>
    29  #include <sys/exec.h>
    30  #include <sys/conf.h>
    31  #include <sys/sysent.h>
    32  #include <sys/lkm.h>
    33  #include <sys/errno.h>
    34  #include <i386/isa/isa_device.h>
    35  #include <i386/isa/isa.h>
    36
    37
</code>

Lines  38 -  51

 -- Declares the device driver entry points as external.
<code>
    38  /*
    39   * declare your entry points as externs
    40   */
    41
    42  extern int pcaprobe(struct isa_device *);
    43  extern int pcaattach(struct isa_device *);
    44  extern int pcaopen(dev_t, int, int, struct proc *);
    45  extern int pcaclose(dev_t, int, int, struct proc *);
    46  extern int pcawrite(dev_t, struct uio *, int);
    47  extern int pcaioctl(dev_t, int, caddr_t);
    48  extern int pcaselect(dev_t, int, struct proc *);
    49  extern void pcaintr(struct clockframe *);
    50  extern struct isa_driver pcadriver;
    51
</code>

Lines  52 -  70

 -- This is creates the device switch entry table for your driver.
This table gets swapped wholesale into the system device switch at
the location specified by your major number.  In the standard model,
these are in /usr/src/sys/i386/i386/conf.c.  NOTE: you cannot pick a
device major number higher than what exists in conf.c, for example at
present, conf.c rev 1.85, there are 67 slots for character devices,
you cannot use a (character) major device number 67 or greater,
without first reserving space in conf.c.
<code>
    52  /*
    53   * build your device switch entry table
    54   */
    55
    56  static struct cdevsw pcacdevsw = {
    57    (d_open_t *)      pcaopen, /* open */
    58    (d_close_t *)     pcaclose, /* close */
    59    (d_rdwr_t *)      enodev, /* read */
    60    (d_rdwr_t *)      pcawrite, /* write */
    61    (d_ioctl_t *)     pcaioctl, /* ioctl */
    62    (d_stop_t *)      enodev, /* stop?? */
    63    (d_reset_t *)     enodev, /* reset */
    64    (d_ttycv_t *)     enodev, /* ttys */
    65    (d_select_t *)    pcaselect, /* select */
    66    (d_mmap_t *)      enodev, /* mmap */
    67    (d_strategy_t *)  enodev /* strategy */
    68  };
    69
    70
</code>

Lines  71 - 131

 -- This section is analogous to the config file declaration of your
device.  The members of the isa_device structure are filled in by what
is known about your device, I/O port, shared memory segment, etc.  We
will probably never have a need for two pcaudio devices in the kernel,
but this example shows how multiple devices can be supported.
<code>
    71  /*
    72   * this lkm arbitrarily supports two 
    73   * instantiations of the pc-audio device.
    74   *
    75   * this is for illustration purposes
    76   * only, it doesn't make much sense
    77   * to have two of these beasts...
    78   */
    79
    80
    81  /*
    82   * these have a direct correlation to the
    83   * config file entries...
    84   */
    85  struct isa_device pcadev[NPCA] = {
    86    {
    87      11,         /* device id */
    88      &amp;pcadriver,  /* driver pointer */
    89      IO_TIMER1,         /* base io address */
    90      -1,      /* interrupt */
    91      -1,         /* dma channel */
    92      (caddr_t)-1,    /* physical io memory */
    93      0,     /* size of io memory */
    94      pcaintr ,       /* interrupt interface */
    95      0,          /* unit number */
    96      0,     /* flags */
    97      0,          /* scsi id */
    98      0,          /* is alive */
    99      0,          /* flags for register_intr */
   100      0,          /* hot eject device support */
   101      1           /* is device enabled */
   102    },
   103  #if NPCA >1
   104    {
   105
   106   /*
   107    * these are all zeros, because it doesn't make
   108    * much sense to be here
   109    * but it may make sense for your device
   110    */
   111
   112      0,         /* device id */
   113      &amp;pcadriver,  /* driver pointer */
   114      0,         /* base io address */
   115      -1,      /* interrupt */
   116      -1,         /* dma channel */
   117      -1,    /* physical io memory */
   118      0,     /* size of io memory */
   119      NULL,       /* interrupt interface */
   120      1,          /* unit number */
   121      0,     /* flags */
   122      0,          /* scsi id */
   123      0,          /* is alive */
   124      0,          /* flags for register_intr */
   125      0,          /* hot eject device support */
   126      1           /* is device enabled */
   127    },
   128  #endif
   129
   130  };
   131
</code>

Lines 132 - 139

 -- This calls the C-preprocessor macro MOD_DEV, which sets up an LKM device
driver, as opposed to an LKM filesystem, or an LKM system call.
<code>
   132  /*
   133   * this macro maps to a function which 
   134   * sets the LKM up for a driver
   135   * as opposed to a filesystem, system call, or misc
   136   * LKM.
   137   */
   138  MOD_DEV("pcaudio_mod", LM_DT_CHAR, 24, &amp;pcacdevsw);
   139
</code>

Lines 140 - 168

 -- This is the function which will be called when the driver is
loaded.  This function tries to work like sys/i386/isa/isa.c
which does the probe/attach calls for a driver at boot time.  The
biggest trick here is that it maps the physical address of the shared
memory segment, which is specified in the isa_device structure to a
kernel virtual address.  Normally the physical address is put in the
config file which builds the isa_device structures in
/usr/src/sys/compile/KERNEL/ioconf.c.  The probe/attach sequence of
/usr/src/sys/isa/isa.c translates the physical address to a virtual
one so that in your probe/attach routines you can do things like 
<verb>
(int *)id->id_maddr = something;
</verb>
and just refer to the shared memory segment via pointers.
<code>
   140  /*
   141   * this function is called when the module is
   142   * loaded; it tries to mimic the behavior
   143   * of the standard probe/attach stuff from
   144   * isa.c
   145   */
   146  int
   147  pcaload(){
   148    int i;
   149    uprintf("PC Audio Driver Loaded\n");
   150    for (i=0; i<NPCA; i++){
   151      /*
   152       * this maps the shared memory address
   153       * from physical to virtual, to be
   154       * consistent with the way
   155       * /usr/src/sys/i386/isa.c handles it.
   156       */
   157      pcadev[i].id_maddr -=0xa0000;
   158      pcadev[i].id_maddr += atdevbase;
   159      if ((*pcadriver.probe)(pcadev+i)) {
   160        (*(pcadriver.attach))(pcadev+i);
   161      } else {
   162        uprintf("PC Audio Probe Failed\n");
   163        return(1);
   164      }
   165    }
   166      return 0;
   167  }
   168
</code>

Lines 169 - 179

 -- This is the function called when your driver is unloaded; it just displays
a message to that effect.
<code>
   169  /*
   170   * this function is called
   171   * when the module is unloaded
   172   */
   173
   174  int
   175  pcaunload(){
   176    uprintf("PC Audio Driver Unloaded\n");
   177    return 0;
   178  }
   179
</code>

Lines 180 - 190

 -- This is the entry point which is specified on the command line of the
modload.  By convention it is named &lt;dev&gt;_mod.  This is how it is
defined in bsd.lkm.mk, the makefile which builds the LKM.  If you name your
module following this convention, you can do ``make load'' and ``make
unload'' from /usr/src/lkm/pcaudio. <p>
Note: this has gone through <em/many/ revisions from release 2.0 to 2.1.
It may or may not be possible to write a module which is portable across
all three releases. <p>
<code>
   180  /*
   181   * this is the entry point specified
   182   * on the modload command line
   183   */
   184
   185  int
   186  pcaudio_mod(struct lkm_table *lkmtp, int cmd, int ver)
   187  {
   188          DISPATCH(lkmtp, cmd, ver, pcaload, pcaunload, nosys);
   189  }
   190
   191  #endif /* NICP > 0 */
</code>

<sect1> Device Type Idiosyncrasies
<sect2> Character
<sect2> Block
<sect2> Network
<sect2> Line Discipline

<sect1> Bus Type Idiosyncrasies
<sect2> ISA
<sect2> EISA
<sect2> PCI
<sect2> SCSI
<sect2> PCCARD

<sect> Kernel Support

<sect1> Data Structures

<sect2> <tt/struct kern_devconf/ Structure
<p>

This structure contains some information about the state of the device
and driver.  It is defined in /usr/src/sys/sys/devconf.h as:
<code>
struct devconf {
        char dc_name[MAXDEVNAME];       /* name */
        char dc_descr[MAXDEVDESCR];     /* description */
        int dc_unit;                    /* unit number */
        int dc_number;                  /* unique id */
        char dc_pname[MAXDEVNAME];      /* name of the parent device */
        int dc_punit;                   /* unit number of the parent */
        int dc_pnumber;                 /* unique id of the parent */
        struct machdep_devconf dc_md;   /* machine-dependent stuff */
        enum dc_state dc_state;         /* state of the device (see above) */
        enum dc_class dc_class;         /* type of device (see above) */
        size_t dc_datalen;              /* length of data */
        char dc_data[1];                /* variable-length data */
};
</code>

<sect2> <tt/struct proc/ Structure
<p>

This structure contains all the information about a process.
It is defined in /usr/src/sys/sys/proc.h:
<code>
/*
 * Description of a process.
 *
 * This structure contains the information needed to manage a thread of
 * control, known in UN*X as a process; it has references to substructures
 * containing descriptions of things that the process uses, but may share
 * with related processes.  The process structure and the substructures
 * are always addressable except for those marked "(PROC ONLY)" below,
 * which might be addressable only on a processor on which the process
 * is running.
 */
struct  proc {
        struct  proc *p_forw;           /* Doubly-linked run/sleep queue. */
        struct  proc *p_back;
        struct  proc *p_next;           /* Linked list of active procs */
        struct  proc **p_prev;          /*    and zombies. */

        /* substructures: */
        struct  pcred *p_cred;          /* Process owner's identity. */
        struct  filedesc *p_fd;         /* Ptr to open files structure. */
        struct  pstats *p_stats;        /* Accounting/statistics (PROC ONLY). */        struct  plimit *p_limit;        /* Process limits. */
        struct  vmspace *p_vmspace;     /* Address space. */
        struct  sigacts *p_sigacts;     /* Signal actions, state (PROC ONLY). */

#define p_ucred         p_cred->pc_ucred
#define p_rlimit        p_limit->pl_rlimit

        int     p_flag;                 /* P_* flags. */
        char    p_stat;                 /* S* process status. */
        char    p_pad1[3];

        pid_t   p_pid;                  /* Process identifier. */
        struct  proc *p_hash;    /* Hashed based on p_pid for kill+exit+... */
        struct  proc *p_pgrpnxt; /* Pointer to next process in process group. */
        struct  proc *p_pptr;    /* Pointer to process structure of parent. */
        struct  proc *p_osptr;   /* Pointer to older sibling processes. */

/* The following fields are all zeroed upon creation in fork. */
#define p_startzero     p_ysptr
        struct  proc *p_ysptr;   /* Pointer to younger siblings. */
        struct  proc *p_cptr;    /* Pointer to youngest living child. */
        pid_t   p_oppid;         /* Save parent pid during ptrace. XXX */
        int     p_dupfd;         /* Sideways return value from fdopen. XXX */

        /* scheduling */
        u_int   p_estcpu;        /* Time averaged value of p_cpticks. */
        int     p_cpticks;       /* Ticks of cpu time. */
        fixpt_t p_pctcpu;        /* %cpu for this process during p_swtime */
        void    *p_wchan;        /* Sleep address. */
        char    *p_wmesg;        /* Reason for sleep. */
        u_int   p_swtime;        /* Time swapped in or out. */
        u_int   p_slptime;       /* Time since last blocked. */

        struct  itimerval p_realtimer;  /* Alarm timer. */
        struct  timeval p_rtime;        /* Real time. */
        u_quad_t p_uticks;              /* Statclock hits in user mode. */
        u_quad_t p_sticks;              /* Statclock hits in system mode. */
        u_quad_t p_iticks;              /* Statclock hits processing intr. */

        int     p_traceflag;            /* Kernel trace points. */
        struct  vnode *p_tracep;        /* Trace to vnode. */

        int     p_siglist;              /* Signals arrived but not delivered. */

        struct  vnode *p_textvp;        /* Vnode of executable. */

        char    p_lock;                 /* Process lock (prevent swap) count. */
        char    p_pad2[3];              /* alignment */

/* End area that is zeroed on creation. */
#define p_endzero       p_startcopy

/* The following fields are all copied upon creation in fork. */
#define p_startcopy     p_sigmask

        sigset_t p_sigmask;     /* Current signal mask. */
        sigset_t p_sigignore;   /* Signals being ignored. */
        sigset_t p_sigcatch;    /* Signals being caught by user. */

        u_char  p_priority;     /* Process priority. */
        u_char  p_usrpri;       /* User-priority based on p_cpu and p_nice. */
        char    p_nice;         /* Process "nice" value. */
        char    p_comm[MAXCOMLEN+1];

        struct  pgrp *p_pgrp;   /* Pointer to process group. */

        struct  sysentvec *p_sysent; /* System call dispatch information. */

        struct  rtprio p_rtprio;        /* Realtime priority. */
/* End area that is copied on creation. */
#define p_endcopy       p_addr
        struct  user *p_addr;   /* Kernel virtual addr of u-area (PROC ONLY). */
        struct  mdproc p_md;    /* Any machine-dependent fields. */

        u_short p_xstat;        /* Exit status for wait; also stop signal. */
        u_short p_acflag;       /* Accounting flags. */
        struct  rusage *p_ru;   /* Exit information. XXX */
};
</code>

<sect2> <tt/struct buf/ Structure
<p>
The <tt/struct buf/ structure is used to interface with the buffer cache.
It is defined in /usr/src/sys/sys/buf.h:

<code>
/*
 * The buffer header describes an I/O operation in the kernel.
 */
struct buf {
        LIST_ENTRY(buf) b_hash;         /* Hash chain. */
        LIST_ENTRY(buf) b_vnbufs;       /* Buffer's associated vnode. */
        TAILQ_ENTRY(buf) b_freelist;    /* Free list position if not active. */
        struct  buf *b_actf, **b_actb;  /* Device driver queue when active. */
        struct  proc *b_proc;           /* Associated proc; NULL if kernel. */
        volatile long   b_flags;        /* B_* flags. */
        int     b_qindex;               /* buffer queue index */
        int     b_error;                /* Errno value. */
        long    b_bufsize;              /* Allocated buffer size. */
        long    b_bcount;               /* Valid bytes in buffer. */
        long    b_resid;                /* Remaining I/O. */
        dev_t   b_dev;                  /* Device associated with buffer. */
        struct {
                caddr_t b_addr;         /* Memory, superblocks, indirect etc. */
        } b_un;
        void    *b_saveaddr;            /* Original b_addr for physio. */
        daddr_t b_lblkno;               /* Logical block number. */
        daddr_t b_blkno;                /* Underlying physical block number. */
                                        /* Function to call upon completion. */
        void    (*b_iodone) __P((struct buf *));
                                        /* For nested b_iodone's. */
        struct  iodone_chain *b_iodone_chain;
        struct  vnode *b_vp;            /* Device vnode. */
        int     b_pfcent;               /* Center page when swapping cluster. */
        int     b_dirtyoff;             /* Offset in buffer of dirty region. */
        int     b_dirtyend;             /* Offset of end of dirty region. */
        struct  ucred *b_rcred;         /* Read credentials reference. */
        struct  ucred *b_wcred;         /* Write credentials reference. */
        int     b_validoff;             /* Offset in buffer of valid region. */
        int     b_validend;             /* Offset of end of valid region. */
        daddr_t b_pblkno;               /* physical block number */
        caddr_t b_savekva;              /* saved kva for transfer while bouncing
 */
        void    *b_driver1;             /* for private use by the driver */
        void    *b_driver2;             /* for private use by the driver */
        void    *b_spc;
        struct  vm_page *b_pages[(MAXPHYS + PAGE_SIZE - 1)/PAGE_SIZE];
        int             b_npages;
};
</code>

<sect2> <tt/struct uio/ Structure
<p>
This structure is used for moving data between the kernel and user spaces
through read() and write() system calls.  It is defined in
/usr/src/sys/sys/uio.h:
<code>
struct uio {
        struct  iovec *uio_iov;
        int     uio_iovcnt;
        off_t   uio_offset;
        int     uio_resid;
        enum    uio_seg uio_segflg;
        enum    uio_rw uio_rw;
        struct  proc *uio_procp;
};

</code>

<sect1> Functions
lots of 'em

<sect> References.

<p> FreeBSD Kernel Sources http://www.freebsd.org
<p> NetBSD Kernel Sources http://www.netbsd.org
<p>  Writing Device Drivers: Tutorial and Reference;
Tim Burke, Mark A. Parenti, Al, Wojtas;
Digital Press, ISBN 1-55558-141-2.

<p> Writing A Unix Device Driver;
Janet I. Egan, Thomas J. Teixeira;
John Wiley &amp; Sons, ISBN 0-471-62859-X.

<p>  Writing Device Drivers for SCO Unix;
Peter Kettle;

</article>