So You Wanna Write Solaris Device Drivers?

James Liu, June 2008


Abstract: This article focuses on providing device drivers for the Solaris OS, beginning with foundational information about the Solaris kernel architecture in general and device drivers in particular. This article then develops a minimal skeleton driver.

Getting Acquainted with Solaris Device Drivers

Whenever a bunch of tech-savvy folks get together and compare operating system kernels, there is a good chance the issue of device support will come up. While the kernel is the core piece of software that bridges the gap between hardware and software, it's the small chunks of code known as device drivers that enable a platform's peripherals to interface with users.

Access by device drivers to and from the kernel relies on a well-defined set of interfaces [1]. The Solaris OS was first released in 1992, so it has a mature set of interfaces for a wide variety of devices. In fact, some developers might have started out writing graphics, network, and storage controller drivers using the Solaris OS years ago. However, until recently, the number of experienced Solaris device driver writers has been a relatively small elite group. Some critics cite the Solaris OS being closed-source as the main reason. But since the release of core kernel code into the open source community in June 2005 [2], the number of programmers engaged in Solaris driver development has grown and attracted a more casual user and developer interest in drivers.

In this article, we get acquainted, or perhaps for some readers, reacquainted with Solaris device drivers. We take a pragmatic approach that looks at device drivers, first from the point of view of a system installer or administrator who wants to configure a binary driver to make it work with a particular device, and second, from the point of view of a developer who wants to port a driver to the Solaris OS or write a driver from scratch for the Solaris OS. We hope to demystify Solaris drivers and make driver development as straightforward as application development.

A First Look at Kernel Modules

For the uninitiated, the first approach to writing a device driver can be very intimidating. You do need to overcome a steep learning curve before you can successfully write device drivers. Part of the obstacle is that device drivers do interact with both the kernel and hardware. The standards for quality are necessarily higher since poor programming of a driver module can kill the whole kernel and panic the entire OS. Applications, however, either in the Solaris OS or in other well architected operating systems, are segmented from the kernel. A poorly written program can consume undue system resources, dump core, or mysteriously exit, but rarely will it bring down the kernel or other independently running programs. Aside from the fact that a device driver is code that runs typically in the kernel, there are many similarities between writing a device driver and writing a standard application. Like applications, device drivers are binary code, usually written in C, and compiled and linked to libraries.

The Solaris Device Tree

The Solaris OS organizes and categorizes drivers into a device hierarchy known as a device tree. We can think of this just like a file directory structure. Just as a file hierarchy has directories, subdirectories, and files, each point in this tree represents a node that might be a branch point, sub-branch point, or endpoint. The device file system, or devfs, is the instantiation of the device tree. In general, parent nodes in the device file system map to bus and bridge controller hardware, and leaf nodes map to other devices and controllers. The drivers for bus and bridge controller hardware are called nexus drivers. Other drivers are called leaf node drivers. Leaf node drivers control hardware that has standalone functionality. Examples of leaf node drivers include network interface cards and RAID host bus adapters. The device file system is discussed in more detail later in this article.

Device tree with leaf nodes and parent nexus nodes.

Figure 1 Sample Device Tree

Types of Device Drivers

Nexus drivers typically control bridge, bus, hub, and switch chips on the motherboard and are usually supplied by the Solaris OS. Nexus drivers are not usually implemented by third party driver writers. The Solaris OS includes nexus drivers for various bus and controller subsystems such as:

  • PCI and PCI-express buses [3]

  • USB nexus

  • Traditional ISA bus controllers for attached keyboards, mice, and serial ports

  • SCSI bus

  • IDE nexus

  • Framework for Generic LAN Devices

Because the Solaris OS supplies the nexus drivers, most developers focus on leaf node drivers, which come in two basic types: character and block. Character device drivers are used to access serial ports and custom I/O boards. Character device drivers can be implemented as non-STREAMS drivers or as STREAMS drivers. (This introduction won't go deeply into STREAMS drivers. Instead, that will be presented in a later article.) Block device drivers are often used to access filesystem disks and storage.

What's important to understand is that device drivers enable the Solaris kernel to access lower level hardware. At a high level, the kernel provides applications with a standard set of APIs that perform networking, device control, process and memory management, and access to storage. However, in order for the kernel to use a particular vendor's device, it needs to have code that provides that low-level functionality. That code is what maps the standard kernel library function calls into actions that will take place with a hardware device. This is the job of the device driver.

The following figure shows an overview of the Solaris OS architecture. Applications are at the user level. Device drivers are in the kernel. The hardware level includes CPU, disks, NICs.

General overview of Solaris OS architecture. Applications are at the user level. Device drivers are in the kernel. The hardware level includes CPU, disks, NICs.

Figure 2 Overview of the Solaris OS Architecture

The Solaris DDI/DKI

Perhaps the main difference between application programming and driver programming is in freedom. The implementation of application programs usually is constrained only by a minimal set of rules that govern how the application interacts with the user's shell or the user's graphical interface. How the application actually does its job can be very flexible and virtually independent of the OS from the developer's perspective. For example, we could write a standalone program that has its own subroutines and we only implement the standard main() entry point, which we know must be implemented in all standalone programs. Of course the program is still interacting and running within the OS, but from our perspective, we are insulated from having to deal with task scheduling, concurrency, memory access, resource allocation, and I/O collisions. Unless we're trying to write a fancy software package, the kernel and its libraries insulate us from dealing with many of these issues.

In driver programming, we have much less freedom and less of a safety net. We are no longer writing standalone code. Instead, we're writing software extensions that will enable the Solaris kernel to access a device. So when the kernel needs to use a device, we need to provide a driver module that implements an agreed-upon set of subroutines to enable the kernel to predictably control that device. Those subroutines follow a strict set of naming and coding constraints. Without those strict rules, the kernel would need to implement custom subroutines for every driver it would possibly load. Implementing custom subroutines in each driver would make the kernel very sensitive to changes in the device, and make it very impractical for an OS to keep up with a growing market for third-party peripheral devices. Today, all operating systems define relatively standardized device driver interfaces (DDIs) and device-kernel interfaces (DKIs).

The Solaris DDI/DKI has two parts. One part provides a stable set of callable function subroutines that simplify development. These functions are also portable, so using a particular DDI/DKI function in a driver allows a Solaris device driver to be compiled and run on a 32-bit x86 platform, on a 64-bit x86 platform, or on a SPARC platform with little or no change to source code. The DDI/DKI will do the right thing when running on the particular platform. The standard C libraries are still available to device driver writers, but when devices need to access I/O and manipulate bits within data, not using the DDI/DKI function calls can lead to big problems with sizes, endianness, and portability between platforms. The following figure shows a block diagram of a typical driver module with basic Solaris DDI/DKI.

Typical driver module with basic Solaris DDI/DKI.

Figure 3 Driver Module Block Diagram

In addition to these callable functions, the other half of the DDI/DKI defines interfaces for callback routines that developers must implement in their drivers. The kernel expects each driver to implement these interfaces and will invoke or “call back” those functions during known phases of the driver module's lifecycle. These callbacks include interfaces needed to handle loading and unloading operations, initializing and finishing, attaching and detaching, setup of interrupt handling, and standard I/O control to the physical device. Finally, developers need to implement additional interfaces in order for the driver to attach to a particular type of nexus controller such as SCSI, SATA, USB, LAN.

A natural question that arises from discussions of driver functions and interfaces is, How and where are driver states managed in the driver? Three primary data structures track driver state in the Solaris OS. Every driver references these three data structures. At the top level is the module operations or mod_ops structure. This structure holds references to a module's state within the Solaris system. We pass the mod_ops structure to the system during initialization and we also pass it to the system during finalization. One of the references within the mod_ops structure points to a second device structure called dev_ops. This device operations structure keeps track of device state and holds pointers to key device callbacks for attach(), detach(), getinfo(), and probe(). The dev_ops structure also contains a reference to the cb_ops structure, which is the actual character or block driver operations structure and contains a list of references to DDI-specific functions implemented in the driver.

Driver Lifecycle

Let's examine the driver lifecycle in a bit more detail. When the kernel loads a driver, it expects the driver to implement certain subroutines and pass in a reference to the module's mod_ops structure to retain state. The required subroutines, defined in the DDI/DKI, at a minimum must implement callbacks for _info() and _init(). Conversely, when the driver is unloaded, the kernel expects to call the driver's _fini() function. We can summarize these first three mandatory callbacks all drivers must implement as:


After the system loads a module (and also when the system is going to unload a module), the kernel looks for one of two device-specific callbacks to do the actual hardware initialization or finalization. These callbacks are similar to those required for the system to manage modules, but these are designed to manage the hardware. We can summarize the next two mandatory callbacks all drivers must implement as:


At a minimum, all device drivers must implement the above five functions as defined in the DDI/DKI. The DDI/DKI also specify some standard functions for communicating with the device and performing I/O to and from the kernel to the device. These functions are:


All of the above callbacks are function names with specific arguments that the Solaris OS requires of drivers so that the kernel knows how to notify and perform I/O with the driver.

You can run the following command to see the signature of the function and arguments, read a synopsis, and possibly see some sample code showing how the function is used within a driver:

                 man -s 9E                      function_name                 

More information can be found in the system documentation that accompanies the Solaris OS. You can also obtain more detailed references online [1,4,5].

System Administration - How the Solaris OS Manages Drivers

This section discusses the device tree, driver management files, and driver management commands.

An Overview of 32-bit and 64-bit Drivers in the Kernel

On x86 platforms, the Solaris OS can run in both 32-bit and 64-bit mode depending on the processor architecture. By default, if we have a 64-bit capable platform, the Solaris OS boots into 64-bit mode and thus requires all its drivers to be 64-bit modules. This is because kernel modules share the same memory address space as the rest of the kernel. If the platform is only 32-bit capable, then the Solaris OS boots into 32-bit mode and loads only 32-bit modules. The general rule is that driver binaries must match the kernel bit-wise in pointer size. However, because applications do not run inside the kernel, a 64-bit kernel can run both 32-bit and 64-bit applications. A 32-bit kernel can only run 32-bit applications.

It is possible to force the boot of the Solaris OS into 32-bit mode on a 64-bit system by editing the grub boot menu during boot time and editing the grub command line. Either edit the file /boot/grub/menu.lst and reboot, or reboot and enter e at the grub menu. Typically, we change references to kernel/amd64/unix or kernel/$ISADIR/unix into kernel/unix. This 32-bit boot mode can be made permanent by editing /boot/grub/menu.lst, adding an appropriate entry for 32-bit, and setting that entry as the default. See [6]. There is no performance reason to boot 32-bit on a 64-bit system. However, we do this to verify that a driver runs equally well in both 32-bit and 64-bit mode on the Solaris OS.

Where We Deploy Driver Modules

When you finish compiling a driver and want to deploy the module, you need to know where the Solaris OS keeps the modules. For x86 and x64 platforms, the Solaris OS expects most driver modules to be in the following directories:


32-bit driver binaries and driver.conf files


64-bit binaries

There are other directories we could copy modules to. The above directories are the recommended places for x86 driver binaries. Typically, when we build a driver module, we generate both 32-bit and 64-bit binaries and deploy them into the directories shown above, even if we only intend to run in one mode or the other. This makes the driver available for both types of platforms if we should ever copy the binaries over to another platform, or boot into a different bit-mode as explained above.

System Driver Interaction and the Device Tree

We mentioned above in The Solaris Device Tree that the Solaris OS actually implements a representation of the device tree as a file system known as devfs. This file system manages the name space of all physical devices known to the Solaris OS and is mounted at boot time under the /devices directory. This isn't a normal file system, and can't be altered using any regular file system commands. Instead, the OS controls this exclusively. Nodes are created in devfs primarily when a driver module attaches to a device during initialization (see Driver Lifecycle). Each node is assigned a major and minor number which are displayed when performing a full directory file listing. The major number represents a particular driver module which the kernel has loaded to interface with this device, and the minor number represents the instance of this device. By default, the first instance number of a device starts with zero (0). If there are additional similar devices that use the same driver module, then they are assigned minor numbers 1, 2, and so on.

While the /devices device file system represents all physical devices, the /dev file system is a logical name mapping to those physical devices. In other words, the logical devices found in /dev are just symbolic links to the actual nodes in /devices. The /dev file system is an actual file system that can be manipulated manually, but this is not recommended.

The reason to have two separate directories for device files is that the same logical device name might be expected by the system or by applications. But on a different platform, that logical device might be serviced by a completely different device from a different manufacturer using a different driver module. Consider the following audio example. If audio is working on a system, we will find /dev/audio and /dev/audioctl nodes as logical devices in /dev. This is the same on any platform with audio. But that /dev/audio might be pointing to an AC'97 audio device on one platform and an HD audio device on another, and each of these devices requires a different type of driver. The same analogy can be applied to disk storage. The Solaris OS might find a number of SCSI disks and a specific target controller. These typically map to /dev/rdsk/cXtYdZsN, where X, Y, Z, and N are numbers that indicate the controller bus, controller target, disk number or LUN, and slice number on that disk. But these logical disk devices might be pointing to an on-board SCSI controller in one case, and in another, a completely different RAID disk Host-Bus Adapter with a RAID volume. Yet the path names follow a similar logical rule, so this eases systems management.

Note that while the contents in /devices can't be manipulated, we can still execute an ls(1) within the directory, which suggests a tip to remember: When testing whether a driver attaches to a device, we can just run a listing and see if a node is actually created in /devices when performing a driver load. We will discuss how we load, unload, and manage the nodes in /devices and /dev in Commands to Manage, Add, Remove, Update and Check Drivers.

Driver Management Files

Up to this point, some readers might still be scratching their heads about how the Solaris OS knows which module to load. Like any other OS, there is a master database file that maps driver modules to the actual device as defined by the Vendor and Device ID extracted either through the PCI bus or by another device hub such as the USB nexus. This text file is located in:


Another text file maps the actual physical device name (actually a path in /devices) to an instance number of a driver name. This file is located in:


The system maps driver names to major numbers in the following file:


The system tracks a database of default permissions or access control lists for each instance of a driver in the following file:


While the driver_aliases, path_to_inst(4), name_to_major, and minor_perms files are hand editable, we recommend that you manipulate them only by using the driver commands discussed in the next section.

Commands to Manage, Add, Remove, Update and Check Drivers

Below is a list of basic commands we use to maintain devices. These commands are the preferred way to manipulate the driver management files discussed above, rather than directly editing those files.


The add_drv(1M) command adds a new driver. For example, to add a driver foobar that will map to a PCI device with root-only permission, we can run the following command:

                     /usr/sbin/add_drv -i '"pci0909,5c"' \
                     -m '* 0600 root sys' foobar

Note that in some cases, a driver might implement or belong to a certain class of device. For example, this is particularly true of storage Host Bus Adapters (HBAs) that implement what looks like a SCSI interface even though we can attach IDE, SATA, or SAS drives to the HBA. To declare to the system that such a driver is exporting the SCSI class interfaces, for example, we simply add -c scsi to the add_drv(1M) command.


The rem_drv(1M) command removes an existing driver. Removes the driver from the /etc/driver_aliases file but might not remove the driver from /kernel/drv until after a reboot if that module is still in use.

                     /usr/sbin/rem_drv foobar

The update_drv(1M) command can update the properties of a driver already known to the system. Often used when adding a new PCI device that is known to use an existing driver. The format of this command is similar to the add_drv command. For example, if we wanted to add a new PCI device to also be supported under the foobar driver, we could run the following command:

                     /usr/sbin/update_drv -i '"pci0909,8f"' \
                     -m '* 0600 root sys' foobar

The modinfo(1M) command displays currently loaded module information. This command is a handy way to check whether a module is loaded into the system and what size the module is.


The modload(1M) command manually loads a module.


The modunload(1M) command tries to unload a module. Doesn't take effect if the driver is currently in use.


The devfsadm(1M) command maintains the /dev links and those in /etc/path_to_inst.

Basics of Building a Driver

Two steps are required when building a driver binary. The first step is to generate a driver binary object. The second step is to link the binary object.

Step 1: Compiling the Binary Object

To generate a driver binary object we can conveniently choose either the Sun Studio compiler [7] or the GNU C compiler ( gcc). Some operating systems such as Linux might require developers to stay within strict guidelines on version of the compiler for a particular kernel version in order for a new or updated kernel module to be binary compatible with the rest of the kernel. Unlike Linux, the Solaris OS maintains strict control over the binary interfaces publicly exported by the kernel and its C libraries. These interfaces do not mutate based on the version of compiler running on any given version of the Solaris OS, so it is usually safe to use any version of either gcc or Sun Studio compilers to compile the driver object. See the appendix in [1] for lists of supported and deprecated interfaces.

While this is not supported or guaranteed, driver binaries compiled on the Solaris OS usually are forward compatible along the same version of the OS. In other words, a Solaris 10 1/06 driver binary should work on the Solaris 10 5/08 OS with no changes, as long as it uses only committed DDI/DKI interfaces. The same applies to Solaris Express or OpenSolaris drivers compiled for earlier versions. These should run with no changes or recompile on forward versions of the OS. Note that the reverse is not true: A driver compiled against a newer OS might not run on an older OS.

We summarize the compiler commands with a few examples below. For a complete summary, see the end of the first chapter of [5].

If you are compiling for a 64-bit x86 architecture using Sun Studio 10 or Sun Studio 11, use both the -xarch=amd64 option and the -xmodel=kernel option:

                 cc -D_KERNEL -xarch=amd64 -xmodel=kernel -c foobar.c

We must use the -D_KERNEL flag to signify that we're compiling for a kernel module. In the case of 64-bit compilation, we need to specify both the architecture ( -xarch=amd64) and the -xmodel flags to insure a properly built 64-bit driver object. Note also that for Studio 12 and later compilers, the -xarch=amd64 is deprecated and the new flag is -m64, just as for gcc. In case you were wondering, yes, we can build 64-bit binaries on a 32-bit running OS. But we need to remember that the default in both 32-bit and 64-bit systems is to build a 32-bit binary in the absence of any particular target architecture flags.

If you are compiling for a 64-bit x86 architecture using the GNU C compiler, use the following compile command:

                 gcc -D_KERNEL -ffreestanding -m64 -c foobar.c

If you are compiling for a 32-bit architecture using the Sun Studio C compiler, use the following compile command:

                 cc -D_KERNEL -c foobar.c

If you are compiling for a 32-bit architecture using the GNU C compiler, use the following compile command:

                 gcc -D_KERNEL -ffreestanding -c foobar.c

Step 2: Linking the Binary Object

All driver binaries must then be linked to resolve symbols in the final binary object. For basic modules that don't depend on any other kernel driver frameworks or other driver modules, the command is simply:

                 ld -r -o foobar foobar.o

In many instances, a driver module also depends on other driver modules or must be dynamically linked to other kernel libraries. We present two examples below.

The following example shows how to link an audio driver binary object that depends on the Solaris OS audio support, mixer, and mixer source modules to generate a binary. The -dy and -N flags are typically used together when dynamically linking driver modules to kernel libraries.

                 ld -r -dy -N misc/audiosup -N misc/mixer \
                 -N misc/amsrc -o audiodrv audiodrv.o

The following example shows how to link a typical network driver with the Solaris OS generic LAN driver (GLD) layer.

                 ld -r -dy -N misc/gld -o nicdrv nicdrv.o

You can easily check whether a given module is 32-bit or 64-bit simply by using the file(1) command on a binary object. The file command should display whether the binary is an ELF 32-bit or ELF 64-bit binary. Lastly, we highly recommend putting these commands into a make(1S) system Makefile to simplify build.

Putting It All Together - A Sample Driver

This section shows a skeleton driver that implements the minimum number of interfaces required. This section then shows the header file for that skeleton driver and the configuration file for that driver.

A Skeleton Driver

Typically, a driver is composed of one or more C files that implement the functions used by the kernel to access the driver. It is usual practice to find the module, device, and driver operations structures declared explicitly in each C file where the operations structure is used and not in the header file. It is common practice to declare the name of the driver as a prefix to functions and global variables that are specific to that driver. For example, the attach() subroutine is implemented as foobar_attach(). Note also that the majority of system interaction to load the module and cause the system to create major and minor nodes is located within the foobar_attach() routine.

 *     foobar.c - example skeleton driver
#include <sys/errno.h>
#include <sys/conf.h>
#include <sys/cmn_err.h>
#include <sys/modctl.h>
#include <sys/sunddi.h>
#include <sys/stat.h>
#include "foobar.h"

static int foobar_attach(dev_info_t *dip, ddi_attach_cmd_t cmd);

static int foobar_detach(dev_info_t *dip, ddi_detach_cmd_t cmd);

static struct cb_ops foobar_cb_ops = {
    nodev,          /* no open */
    nodev,          /* no close */
    nodev,          /* no strategy (only for block drivers) */
    nodev,          /* no print */
    nodev,          /* no dump (only for block drivers) */
    nodev,          /* no read */
    nodev,          /* no write */
    nodev,          /* no ioctl */
    nodev,          /* no devmap */
    nodev,          /* no mmap */
    nodev,          /* no segmap */
    nochpoll,       /* no chpoll entry point */
    ddi_prop_op,    /* Use system-supplied prop_op entry point */
    D_NEW | D_MP

static struct dev_ops foobar_ops = {
    0,/* reference count: always 0 initially */
    nulldev,                  /* No getinfo entry point */
    nulldev,                  /* DEPRECATED: identify entry point */
    nulldev,                  /* no probe entry point */
    nodev,                    /* no reset entry point */
    &foobar_cb_ops,           /* Reference the cb_ops defined above */
    (struct bus_ops *)NULL    /* Not a nexus driver, so no bus_ops */

extern struct mod_ops mod_driverops;

static struct modldrv Modldrv = {
        &mod_driverops,    /* Use system-supplied mod_driverops */
        "foobar driver v" FOOBAR_VERSION, /* Module Name/Version */

static struct modlinkage Modlinkage = {

 * This bit of static data is used by the DDI to
 * keep track of the per-instance driver "soft state"
static void *soft_statep;
     * Initialize the soft state APIs so we can
     * allocate soft state in foobar_attach()
    if (ddi_soft_state_init(&soft_statep,
        sizeof (struct foobar_state), 1)
        != DDI_SUCCESS)
            return (DDI_FAILURE);

    if (mod_install(&Modlinkage) != 0) {
        return (-1);

    return (0);

_info(struct modinfo *modinfop)
    return (mod_info(&Modlinkage, modinfop));

    return (mod_remove(&Modlinkage));

static int
foobar_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
    /* Use the instance number as the minor number */
    instance = ddi_get_instance(dip);

    if (ddi_soft_state_zalloc(soft_statep, instance)
        == DDI_FAILURE)
        return (DDI_FAILURE);

    softp = ddi_get_soft_state(soft_statep, instance);
    ASSERT(softp != NULL);

    if (ddi_create_minor_node(dip, FOOBAR_MINOR_NAME,
        S_IFCHR, instance, DDI_PSEUDO, 0) 
        != DDI_SUCCESS) {
        cmn_err(CE_WARN, "Minor creation failed!");
        return (DDI_FAILURE);
    softp->init_state |= FOOBAR_INIT_MINOR;
    softp->dip = dip;
    mutex_init(&softp->mutex, NULL, MUTEX_DRIVER, 0);
    softp->buffer = (char *)kmem_alloc(FOOBAR_BUFLEN,
    ddi_report_dev(dip);    /* Announce we've attached! */
    return (DDI_SUCCESS);

static int
foobar_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
    int instance;
    struct foobar_state *softp;

    if (cmd != DDI_DETACH)
        return (DDI_FAILURE);

    /* Use the instance number as the minor number */
    instance = ddi_get_instance(dip);

    softp = ddi_get_soft_state(soft_statep, instance);

    ASSERT(softp != NULL);
    if (softp->init_state & FOOBAR_INIT_MINOR) {
        /* Remove minor nodes associated with dip */
        ddi_remove_minor_node(dip, NULL);
    ASSERT(softp->buffer != NULL);
    kmem_free(softp->buffer, FOOBAR_BUFLEN);

    ddi_soft_state_free(soft_statep, instance);

    return (DDI_SUCCESS);

The Skeleton Driver Header File

The header file contains remaining declarations such as those defined below.

 *      foobar.h - example skeleton driver header

#ifndef _FOOBAR_H
#define _FOOBAR_H

#define FOOBAR_INIT_MINOR 0x00000001

#define FOOBAR_VERSION "1.0"
#define FOOBAR_BUFLEN 1024
#define FOOBAR_MINOR_NAME "xyzzy"

struct foobar_state {
    dev_info_t    *dip;          /* Opaque dev_info pointer */
    int           init_state;    /* See FOOBAR_INIT_* */
    kmutex_t      mutex;         /* driver lock */
    char          *buffer;       /* message buffer */

#endif /* #ifdef _FOOBAR_H */

The Skeleton Driver Configuration File

The last file is the driver configuration file. In this example, the driver configuration file is foobar.conf. The driver .conf file is used to override driver internal settings and is read once during driver initialization. In most modern drivers, it is not necessary to have a .conf file, since these can be explicitly set inside the driver code itself. The format of this file is usually to specify a target driver name and the driver's parent node, as shown below.

# foobar.conf - example skeleton driver conf file

name="foobar" parent="pseudo" instance=0;

Recall that this driver.conf file should be placed in the 32-bit driver directory ( /kernel/drv) even if the system is 64-bit.

What's Next?

In the next article in this series, we'll cover some more frameworks for specific types of drivers people write and provide some insights into when to use which framework for a driver. Meanwhile, developers should have a peak at the web site and consider joining the device driver community [8].


This paper would not be possible without the help and contributions of kernel expert Seth Goldberg, who works at Sun Microsystems. This document is based on work that Seth coauthored with James and delivered at the Intel Developer Forum in 2007 on Solaris Device Drivers.


  1. Writing Device Drivers,, Sun Microsystems, Inc., 2008.

  2. source tree,

  3. PCI SIG: PCI documentation and standards,

  4. Max Bruning, “Inside OpenSolaris: Introduction to Solaris Drivers,”, March 31, 2005.

  5. Device Driver Tutorial,, Sun Microsystems, Inc., 2008.

  6. “Boot into 32-bit kernel on 64-bit platform,”, January 30, 2008.

  7. OpenSolaris Device Drivers Community,

Comments (latest comments first)

Discuss and comment on this resource in the BigAdmin Wiki

Unless otherwise licensed, code in all technical manuals herein (including articles, FAQs, samples) is provided under this License.

Left Curve
Popular Downloads
Right Curve
Untitled Document
Left Curve
More Systems Downloads
Right Curve