Release Notes for XFree86[tm] 4.0 : X server
Previous: Introduction
Next: X libraries and clients

2. X server

Unlike XFree86 3.3.x where there are multiple X server binaries, each of which drive different hardware, XFree86 4.0 has a single X server binary called XFree86. This binary can either have one or more video drivers linked in statically, or, more usually, dynamically load the video drivers and other modules that are needed.

XFree86 4.0 has X server support for most UNIX(R) and UNIX-like operating systems on Intel/x86 platforms, plus support for Linux on Alpha and PowerPC platforms. Work on support for additional architectures and operating systems is in progress, and is planned for future releases.

2.1. Loader and Modules

The XFree86 X server has a built-in run-time loader, donated by Metro Link. This loader can load normal object files and libraries in most of the commonly used formats. Since the loader doesn't rely on an operating system's native dynamic loader support, it works on platforms that don't provide this feature, and makes it possible for the modules to be operating system independent (although not, of course, independent of CPU architecture). This means that, for example, a module compiled on Linux/x86 can be loaded by an X server running on Solaris/x86, or FreeBSD, or even OS/2. One of the main benefits of this is that when modules are updated, they don't need to be recompiled for each different operating system. We're planning to take advantage of this to provide more frequent driver module updates in between major releases.

The loader in version 4.0 has support for Intel (x86), Alpha and PowerPC platforms. It also has preliminary support for Sparc platforms but this isn't used yet.

The X server makes use of modules for video drivers, X server extensions, font rasterisers, input device drivers, framebuffer layers (like mfb, cfb, etc), and internal components used by some drivers (like XAA),

The module interfaces (API and ABI) used in this release is still subject to change without notice. While we'll attempt to provide backward compatibility for the module interfaces as of the 4.0 release (meaning that 4.0 modules will work with future core X server binaries), we can't guarantee that this will be the case. We are planning to fully document and stabilise the module interfaces in a future release, and at that point backward compatibility will be easier to achieve.

Note about module security

The XFree86 X server runs with root privileges, which means that the X server loadable modules also run with these privileges. For this reason we recommend that all users be careful to only use loadable modules from reliable sources. We hope to have a mechanism for signing/verifying the modules that we provide available in a future release.

2.2. Configuration File

The X server configuration file format has been extended to handle some of the new functionality. The xf86config utility can be used to generate a basic config file, that may require some manual editing. The X server also has preliminary support for generating a basic config file. This is done by running (as root) "XFree86 -configure". Alternatively, the sample config file that is installed in /usr/X11R6/lib/X11 may be used as a starting point. The XF86Setup utility is currently not usable, but work is continuing in this area.

The main changes are covered here, but please refer to the XF86Config manual page for more comprehensive information:

The config file search patch has been extended, with the directories /etc/X11 and /usr/X11R6/etc/X11 being added. The full search path details are documented in the XF86Config manual page.

2.3. Command Line Options

The following new X server command line options have been added:

-depth n

This specifies the colour depth that the server is running at. The default is 8 for most drivers. Most drivers support the values 8, 15, 16 and 24. Some drivers also support the values 1 and 4. Some drivers may also support other depths. Note that the depth is different from the ``bpp'' that was specified with previous versions. The depth is the number of bits in each pixel that are significant in determining the pixel's value. The bpp is the total size occupied by each pixel, including bits that are not used. The old -bpp option is no longer recognised because it isn't a good way of specifying the server behaviour.

-fbbpp n

This specifies the bpp format to use for the framebuffer. This may be used in 24-bit mode to force a framebuffer format that is different from what the driver chooses by default. In most cases there should be no need to use this option.


This specifies that the client-side pixmap format should be the packed 24-bit format that was often used by the 3.3.x servers. The default is the more common 32-bit format. There should normally be no need to use this option.


This specifies that the client-side pixmap format should be the sparse 32-bit format. This is the default, so there should normally be no need to use this option.

-layout name

This specifies which ServerLayout section in the config file to use. When this option is not specified, the first ServerLayout section is used. When there is no ServerLayout section, the first Screen section is used.

-screen name

This specifies which Screen section in the config file to use. When this option is not specified, the first ServerLayout section is used. When there is no ServerLayout section, the first Screen section is used.

-keyboard name

This specifies which InputDevice section in the config file to use for the core keyboard. This option may be used in conjunction with the -screen option.

-pointer name

This specifies which InputDevice section in the config file to use for the core pointer. This option may be used in conjunction with the -screen option.

-modulepath path

This specifies the module search path. The path should be a comma-separated list of absolute directory paths to search for server modules. When specified here, it overrides the value specified in the config file. This option is only available when the server is started by the root user.

-logfile file

This specifies the log file name. When specified here, it overrides the default value. This option is only available when the server is started by the root user.


This specifies that the scanpci module should be loaded and executed. This does a scan of the PCI bus.

-logverbose [n]

This options specifies the verbosity level to use for the log file. The default is 3.

The following X server command line options have been changed since 3.3.x:

-verbose [n]

This option specifies the verbosity level to use for the server messages that get written to stderr. It may be specified multiple times to increase the verbosity level (as with 3.3.x), or the verbosity level may be specified explicitly as a number. The default verbosity level is 1.

-xf86config filename

This option has been extended to allow non-root users to specify a relative config file name. The config file search path will be used to locate the file in this case. This makes it possible for users to choose from multiple config files that the the sysadmin has provided.

2.4. XAA

The XFree86 Acceleration Architecture (XAA) has been completely rewritten from scratch. Most drivers implement acceleration by making use of the XAA module.

2.5. Multi-head

Some multi-head configurations are supported in this release, primarily with multiple PCI/AGP cards. However, this is an area that is still being worked on, and we expect that the range of configurations for which it works well will increase in future releases. A configuration that is known to work well in most cases is multiple (supported) Matrox cards.

One of the main problems is with drivers not sufficiently initialising cards that were not initialised at boot time. This has been improved somewhat with the INT10 support that is used by most drivers (which allows secondary card to be "soft-booted", but in some cases there are other issues that still need to be resolved. Some combinations can be made to work better by changing which card is the primary card (either by using a different PCI slot, or by changing the system BIOS's preference for the primary card).

2.6. Xinerama

Xinerama is an X server extension that allows multiple physical screens to behave as a single screen. With traditional multi-head in X11, windows cannot span or cross physical screens. Xinerama removes this limitation. Xinerama does, however, require that the physical screens all have the same root depth, so it isn't possible, for example, to use an 8-bit screen together with a 16-bit screen in Xinerama mode.

Xinerama is not enabled by default, and can be enabled with the +xinerama command line option for the X server.

Xinerama was included with X11R6.4. The version included in this release was completely rewritten for improved performance and correctness.

Known problems:

2.7. XVideo extension

The XVideo extension is included in this release, but nobody seems interested in writing up some information about it.

2.8. DGA version 2

DGA 2.0 is nearly completed but still not implemented by all drivers. Preliminary documentation for the client libraries can be found in the xc/programs/Xserver/hw/xfree86/DGA document. Some degree of backwards compatibility with version 1.0 is provided. This information is out of date.

2.9. DDC

The VESA(R) Display Data Channel (DDC[tm]) standard allows the monitor to tell the video card (or on some cases the computer directly) about itself; particularly the supported screen resolutions and refresh rates.

Partial or complete DDC support is available in most of the video drivers. DDC is enabled by default, but can be disabled with a "Device" section entry: Option "NoDDC". We have support for DDC versions 1 and 2; these can be disabled independently with Option "NoDDC1" and Option "NoDDC2".

At startup the server prints out DDC information from the display, but it does not yet use it the determine modelines. For some drivers, the X server's new -configure option uses the DDC information when generating the config file.

Changed behavior caused by DDC. Several drivers uses DDC information to set the screen size and pitch. This can be overridden by explicitly resetting it to the and non-DDC default value 75 with the -dpi 75 command line option for the X server, or by specifying appropriate screen dimensions with the "DisplaySize" keyword in the "Monitor" section of the config file.

2.10. GLX and the Direct Rendering Infrastructure (DRI)

Precision Insight has been provided with funding and support from Red Hat, SGI, 3Dfx, Intel, ATI, and Matrox to integrate the GLX extension for 3D rendering in an X11 window. The 3D core rendering component is the Mesa library. SGI has released the sources to the GLX extension framework under an open license, which essentially provides the glue between the 3D library and this windowing system. Precision Insight has integrated these components into the XFree86 X Server and added a Direct Rendering Infrastructure (DRI). Direct Rendering provides a highly optimized path for sending 3D data directly to the graphics hardware. This release provides a complete implementation of direct rendering support for the 3Dfx Banshee and Voodoo3 graphics cards. Additional direct rendering drivers will be available for 3Dfx, Intel, ATI and Matrox boards during the second quarter of 2000. Updated information on DRI compatible drivers can be found at the DRI Project on SourceForge.

2.11. X-Video Extension (Xv)

An XvQueryPortAttributes function has been added as well as support for XvImages. XvImages are XImages in alternate color spaces such as YUV and can be passed to the server through shared memory segments. This allows clients to display YUV data with high quality hardware scaling and filtering. XvImages are only supported by the Matrox G200/G400 cards at the moment.

2.12. Other extensions

The XFree86-Misc extension has not been fully ported to the new server architecture yet. This should be completed in a future release.

The XFree86-VidModeExtension extension has been updated, and mostly ported to the new server architecture. The area of mode validation needs further work, and the extension should be used with care. This extension has support for changing the gamma setting at run-time, for modes where this is possible. The new xgamma utility makes use of this feature. Compatibility with the 3.3.x version of the extension is provided. The missing parts of this extension and some new features should be completed in a future release.

2.13. Drivers

XFree86 4.0 includes the following drivers:

Driver NameDescription
apmAlliance Pro Motion
chipsChips & Technologies
cirrusCirrus Logic
cyrix (*)Cyrix MediaGX
fbdevLinux fbdev
glideGlide2x (3Dfx)
glint3Dlabs, TI
i740Intel i740
i810Intel i810
r128ATI Rage 128
s3virgeS3 ViRGE
tsengTseng Labs
vgaGeneric VGA

Drivers marked with (*) are present in a preliminary form in this release, but are not complete and/or stable yet.

2.13.1. APM

This is the driver for Alliance AT3D/AT25 and AT24 chips. There is a rather complete support for the functions with acceleration at 8,15,16,24 and 32 bits (limited by the chip at 24bpp). There is preliminary, still buggy, support for the AP6422 chip, which is still supported in 3.3.x servers. The Xv driver is almost ok. The Rush extension for glide2x works, with some additions, including overlay of the result. DGA and DGA2 have been tested ok. Further information can be found in README.apm.

2.13.2. Chips & Technologies

Information about the C&T driver can be found in README.chips.

2.13.3. s3virge

The s3virge driver is a port of the 3.3.x SVGA S3 ViRGE driver. As such it should be as stable and functional as previous XFree86 releases. There are a couple additional benefits included primarily due to common enhancements:

Outstanding items not implemented or fully tested:

Further information can be found in README.s3virge.

2.13.4. TGA

The TGA driver is now accelerated and supports both 8 and 32 plane framebuffers. It is known to work under Linux/Alpha. Please see the README.DECtga file for further information.

2.13.5. Matrox

The MGA driver supports the same range or hardware as XFree86 3.3.4, but has a number of enhancements including multi-head support and support for (non-destructive) overlays (8-bit + 24-bit).

Option "overlay" when the server is started in 32bpp (-fbbpp 32) will enable the 8+24 mode. The current implementation doesn't optimize away unnecessary exposures yet so the performance of this option will be better in future release. By default, the color key for the overlays is 255, but this can be changed with the "ColorKey" option to work around problems in specific programs. Valid values for the key are 2-255.

This release contains performance enhancements for the G400 and particularly for the G400 MAX. It also includes XvImage support for G200/G400 chips and improved memory autodetection support.

Further information can be found in the mga man page.

2.13.6. ATI

Information about the ATI driver can be found in README.ati. The current version is not accelerated. Acceleration support is planned for a future release.

2.13.7. NVIDIA

The "nv" driver supports all Riva TNT accelerators as well as the new GeForce and Quadro accelerators. DGA 2.0 support is included.

Further information can be found in the nv man page.

2.13.8. Glide

This driver is for Voodoo 1 and Voodoo 2 boards. It runs X on top of the 3DFX Glide API (where this is available, like for Linux). You need to have Glide 2.x installed before you can run this driver. This driver uses no hardware acceleration (since there is no 2D acceleration in these boards) but is rather quick anyway since the CPU renders to local RAM which is then copied block-wise to the board. Unfortunately the Voodoo 1/2 boards are rather limited in resolution. The Voodoo 1 can do 800x600 and the Voodoo 2 can do 1024x768 at best, but still it has some use as a second head in Xinerama or multihead mode.

16 and 24 bpp modes are supported (24 bit in 32-bit sparse-packed mode).

Further information about this driver can be found in the 'glide' driver man page for XFree86. You will not get this driver running before reading this man page.

For Voodoo Banshee and Voodoo 3 boards or later: Please use the tdfx driver which talks directly to the hardware and is much faster.

2.13.9. GLINT

The "glint" driver supports most 3Dlabs/Texas Instruments GLINT/Permedia chips. There is a rather complete support (better than in 3.3.x) for acceleration at 8, 15, 16, and 24 bit depths (limited by some chips at some depths). 8+24 overlay is supported. The Xv extension is supported for some boards.

Further information about this driver can be found in the 'glint' driver man page.

Release Notes for XFree86[tm] 4.0 : X server
Previous: Introduction
Next: X libraries and clients