<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN"
	"http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []>

<book id="drmDevelopersGuide">
  <bookinfo>
    <title>Linux DRM Developer's Guide</title>

    <copyright>
      <year>2008-2009</year>
      <holder>
	Intel Corporation (Jesse Barnes &lt;jesse.barnes@intel.com&gt;)
      </holder>
    </copyright>

    <legalnotice>
      <para>
	The contents of this file may be used under the terms of the GNU
	General Public License version 2 (the "GPL") as distributed in
	the kernel source COPYING file.
      </para>
    </legalnotice>
  </bookinfo>

<toc></toc>

  <!-- Introduction -->

  <chapter id="drmIntroduction">
    <title>Introduction</title>
    <para>
      The Linux DRM layer contains code intended to support the needs
      of complex graphics devices, usually containing programmable
      pipelines well suited to 3D graphics acceleration.  Graphics
      drivers in the kernel may make use of DRM functions to make
      tasks like memory management, interrupt handling and DMA easier,
      and provide a uniform interface to applications.
    </para>
    <para>
      A note on versions: this guide covers features found in the DRM
      tree, including the TTM memory manager, output configuration and
      mode setting, and the new vblank internals, in addition to all
      the regular features found in current kernels.
    </para>
    <para>
      [Insert diagram of typical DRM stack here]
    </para>
  </chapter>

  <!-- Internals -->

  <chapter id="drmInternals">
    <title>DRM Internals</title>
    <para>
      This chapter documents DRM internals relevant to driver authors
      and developers working to add support for the latest features to
      existing drivers.
    </para>
    <para>
      First, we go over some typical driver initialization
      requirements, like setting up command buffers, creating an
      initial output configuration, and initializing core services.
      Subsequent sections cover core internals in more detail,
      providing implementation notes and examples.
    </para>
    <para>
      The DRM layer provides several services to graphics drivers,
      many of them driven by the application interfaces it provides
      through libdrm, the library that wraps most of the DRM ioctls.
      These include vblank event handling, memory
      management, output management, framebuffer management, command
      submission &amp; fencing, suspend/resume support, and DMA
      services.
    </para>
    <para>
      The core of every DRM driver is struct drm_driver.  Drivers
      typically statically initialize a drm_driver structure,
      then pass it to drm_init() at load time.
    </para>

  <!-- Internals: driver init -->

  <sect1>
    <title>Driver initialization</title>
    <para>
      Before calling the DRM initialization routines, the driver must
      first create and fill out a struct drm_driver structure.
    </para>
    <programlisting>
      static struct drm_driver driver = {
	/* Don't use MTRRs here; the Xserver or userspace app should
	 * deal with them for Intel hardware.
	 */
	.driver_features =
	    DRIVER_USE_AGP | DRIVER_REQUIRE_AGP |
	    DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_MODESET,
	.load = i915_driver_load,
	.unload = i915_driver_unload,
	.firstopen = i915_driver_firstopen,
	.lastclose = i915_driver_lastclose,
	.preclose = i915_driver_preclose,
	.save = i915_save,
	.restore = i915_restore,
	.device_is_agp = i915_driver_device_is_agp,
	.get_vblank_counter = i915_get_vblank_counter,
	.enable_vblank = i915_enable_vblank,
	.disable_vblank = i915_disable_vblank,
	.irq_preinstall = i915_driver_irq_preinstall,
	.irq_postinstall = i915_driver_irq_postinstall,
	.irq_uninstall = i915_driver_irq_uninstall,
	.irq_handler = i915_driver_irq_handler,
	.reclaim_buffers = drm_core_reclaim_buffers,
	.get_map_ofs = drm_core_get_map_ofs,
	.get_reg_ofs = drm_core_get_reg_ofs,
	.fb_probe = intelfb_probe,
	.fb_remove = intelfb_remove,
	.fb_resize = intelfb_resize,
	.master_create = i915_master_create,
	.master_destroy = i915_master_destroy,
#if defined(CONFIG_DEBUG_FS)
	.debugfs_init = i915_debugfs_init,
	.debugfs_cleanup = i915_debugfs_cleanup,
#endif
	.gem_init_object = i915_gem_init_object,
	.gem_free_object = i915_gem_free_object,
	.gem_vm_ops = &amp;i915_gem_vm_ops,
	.ioctls = i915_ioctls,
	.fops = {
		.owner = THIS_MODULE,
		.open = drm_open,
		.release = drm_release,
		.ioctl = drm_ioctl,
		.mmap = drm_mmap,
		.poll = drm_poll,
		.fasync = drm_fasync,
#ifdef CONFIG_COMPAT
		.compat_ioctl = i915_compat_ioctl,
#endif
		.llseek = noop_llseek,
		},
	.pci_driver = {
		.name = DRIVER_NAME,
		.id_table = pciidlist,
		.probe = probe,
		.remove = __devexit_p(drm_cleanup_pci),
		},
	.name = DRIVER_NAME,
	.desc = DRIVER_DESC,
	.date = DRIVER_DATE,
	.major = DRIVER_MAJOR,
	.minor = DRIVER_MINOR,
	.patchlevel = DRIVER_PATCHLEVEL,
      };
    </programlisting>
    <para>
      In the example above, taken from the i915 DRM driver, the driver
      sets several flags indicating what core features it supports;
      we go over the individual callbacks in later sections.  Since
      flags indicate which features your driver supports to the DRM
      core, you need to set most of them prior to calling drm_init().  Some,
      like DRIVER_MODESET can be set later based on user supplied parameters,
      but that's the exception rather than the rule.
    </para>
    <variablelist>
      <title>Driver flags</title>
      <varlistentry>
	<term>DRIVER_USE_AGP</term>
	<listitem><para>
	    Driver uses AGP interface
	</para></listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_REQUIRE_AGP</term>
	<listitem><para>
	    Driver needs AGP interface to function.
	</para></listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_USE_MTRR</term>
	<listitem>
	  <para>
	    Driver uses MTRR interface for mapping memory.  Deprecated.
	  </para>
	</listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_PCI_DMA</term>
	<listitem><para>
	    Driver is capable of PCI DMA.  Deprecated.
	</para></listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_SG</term>
	<listitem><para>
	    Driver can perform scatter/gather DMA.  Deprecated.
	</para></listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_HAVE_DMA</term>
	<listitem><para>Driver supports DMA.  Deprecated.</para></listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_HAVE_IRQ</term><term>DRIVER_IRQ_SHARED</term>
	<listitem>
	  <para>
	    DRIVER_HAVE_IRQ indicates whether the driver has an IRQ
	    handler.  DRIVER_IRQ_SHARED indicates whether the device &amp;
	    handler support shared IRQs (note that this is required of
	    PCI drivers).
	  </para>
	</listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_DMA_QUEUE</term>
	<listitem>
	  <para>
	    Should be set if the driver queues DMA requests and completes them
	    asynchronously.  Deprecated.
	  </para>
	</listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_FB_DMA</term>
	<listitem>
	  <para>
	    Driver supports DMA to/from the framebuffer.  Deprecated.
	  </para>
	</listitem>
      </varlistentry>
      <varlistentry>
	<term>DRIVER_MODESET</term>
	<listitem>
	  <para>
	    Driver supports mode setting interfaces.
	  </para>
	</listitem>
      </varlistentry>
    </variablelist>
    <para>
      In this specific case, the driver requires AGP and supports
      IRQs.  DMA, as discussed later, is handled by device-specific ioctls
      in this case.  It also supports the kernel mode setting APIs, though
      unlike in the actual i915 driver source, this example unconditionally
      exports KMS capability.
    </para>
  </sect1>

  <!-- Internals: driver load -->

  <sect1>
    <title>Driver load</title>
    <para>
      In the previous section, we saw what a typical drm_driver
      structure might look like.  One of the more important fields in
      the structure is the hook for the load function.
    </para>
    <programlisting>
      static struct drm_driver driver = {
      	...
      	.load = i915_driver_load,
        ...
      };
    </programlisting>
    <para>
      The load function has many responsibilities: allocating a driver
      private structure, specifying supported performance counters,
      configuring the device (e.g. mapping registers &amp; command
      buffers), initializing the memory manager, and setting up the
      initial output configuration.
    </para>
    <para>
      If compatibility is a concern (e.g. with drivers converted over
      to the new interfaces from the old ones), care must be taken to
      prevent device initialization and control that is incompatible with
      currently active userspace drivers.  For instance, if user
      level mode setting drivers are in use, it would be problematic
      to perform output discovery &amp; configuration at load time.
      Likewise, if user-level drivers unaware of memory management are
      in use, memory management and command buffer setup may need to
      be omitted.  These requirements are driver-specific, and care
      needs to be taken to keep both old and new applications and
      libraries working.  The i915 driver supports the "modeset"
      module parameter to control whether advanced features are
      enabled at load time or in legacy fashion.
    </para>

    <sect2>
      <title>Driver private &amp; performance counters</title>
      <para>
	The driver private hangs off the main drm_device structure and
	can be used for tracking various device-specific bits of
	information, like register offsets, command buffer status,
	register state for suspend/resume, etc.  At load time, a
	driver may simply allocate one and set drm_device.dev_priv
	appropriately; it should be freed and drm_device.dev_priv set
	to NULL when the driver is unloaded.
      </para>
      <para>
	The DRM supports several counters which may be used for rough
	performance characterization.  Note that the DRM stat counter
	system is not often used by applications, and supporting
	additional counters is completely optional.
      </para>
      <para>
	These interfaces are deprecated and should not be used.  If performance
	monitoring is desired, the developer should investigate and
	potentially enhance the kernel perf and tracing infrastructure to export
	GPU related performance information for consumption by performance
	monitoring tools and applications.
      </para>
    </sect2>

    <sect2>
      <title>Configuring the device</title>
      <para>
	Obviously, device configuration is device-specific.
	However, there are several common operations: finding a
	device's PCI resources, mapping them, and potentially setting
	up an IRQ handler.
      </para>
      <para>
	Finding &amp; mapping resources is fairly straightforward.  The
	DRM wrapper functions, drm_get_resource_start() and
	drm_get_resource_len(), may be used to find BARs on the given
	drm_device struct.  Once those values have been retrieved, the
	driver load function can call drm_addmap() to create a new
	mapping for the BAR in question.  Note that you probably want a
	drm_local_map_t in your driver private structure to track any
	mappings you create.
<!-- !Fdrivers/gpu/drm/drm_bufs.c drm_get_resource_* -->
<!-- !Finclude/drm/drmP.h drm_local_map_t -->
      </para>
      <para>
	if compatibility with other operating systems isn't a concern
	(DRM drivers can run under various BSD variants and OpenSolaris),
	native Linux calls may be used for the above, e.g. pci_resource_*
	and iomap*/iounmap.  See the Linux device driver book for more
	info.
      </para>
      <para>
	Once you have a register map, you may use the DRM_READn() and
	DRM_WRITEn() macros to access the registers on your device, or
	use driver-specific versions to offset into your MMIO space
	relative to a driver-specific base pointer (see I915_READ for
	an example).
      </para>
      <para>
	If your device supports interrupt generation, you may want to
	set up an interrupt handler when the driver is loaded.  This
	is done using the drm_irq_install() function.  If your device
	supports vertical blank interrupts, it should call
	drm_vblank_init() to initialize the core vblank handling code before
	enabling interrupts on your device.  This ensures the vblank related
	structures are allocated and allows the core to handle vblank events.
      </para>
<!--!Fdrivers/char/drm/drm_irq.c drm_irq_install-->
      <para>
	Once your interrupt handler is registered (it uses your
	drm_driver.irq_handler as the actual interrupt handling
	function), you can safely enable interrupts on your device,
	assuming any other state your interrupt handler uses is also
	initialized.
      </para>
      <para>
	Another task that may be necessary during configuration is
	mapping the video BIOS.  On many devices, the VBIOS describes
	device configuration, LCD panel timings (if any), and contains
	flags indicating device state.  Mapping the BIOS can be done
	using the pci_map_rom() call, a convenience function that
	takes care of mapping the actual ROM, whether it has been
	shadowed into memory (typically at address 0xc0000) or exists
	on the PCI device in the ROM BAR.  Note that after the ROM
	has been mapped and any necessary information has been extracted,
	it should be unmapped; on many devices, the ROM address decoder is
	shared with other BARs, so leaving it mapped could cause
	undesired behavior like hangs or memory corruption.
<!--!Fdrivers/pci/rom.c pci_map_rom-->
      </para>
    </sect2>

    <sect2>
      <title>Memory manager initialization</title>
      <para>
	In order to allocate command buffers, cursor memory, scanout
	buffers, etc., as well as support the latest features provided
	by packages like Mesa and the X.Org X server, your driver
	should support a memory manager.
      </para>
      <para>
	If your driver supports memory management (it should!), you
	need to set that up at load time as well.  How you initialize
	it depends on which memory manager you're using: TTM or GEM.
      </para>
      <sect3>
	<title>TTM initialization</title>
	<para>
	  TTM (for Translation Table Manager) manages video memory and
	  aperture space for graphics devices. TTM supports both UMA devices
	  and devices with dedicated video RAM (VRAM), i.e. most discrete
	  graphics devices.  If your device has dedicated RAM, supporting
	  TTM is desirable.  TTM also integrates tightly with your
	  driver-specific buffer execution function.  See the radeon
	  driver for examples.
	</para>
	<para>
	  The core TTM structure is the ttm_bo_driver struct.  It contains
	  several fields with function pointers for initializing the TTM,
	  allocating and freeing memory, waiting for command completion
	  and fence synchronization, and memory migration.  See the
	  radeon_ttm.c file for an example of usage.
	</para>
	<para>
	  The ttm_global_reference structure is made up of several fields:
	</para>
	<programlisting>
	  struct ttm_global_reference {
	  	enum ttm_global_types global_type;
	  	size_t size;
	  	void *object;
	  	int (*init) (struct ttm_global_reference *);
	  	void (*release) (struct ttm_global_reference *);
	  };
	</programlisting>
	<para>
	  There should be one global reference structure for your memory
	  manager as a whole, and there will be others for each object
	  created by the memory manager at runtime.  Your global TTM should
	  have a type of TTM_GLOBAL_TTM_MEM.  The size field for the global
	  object should be sizeof(struct ttm_mem_global), and the init and
	  release hooks should point at your driver-specific init and
	  release routines, which probably eventually call
	  ttm_mem_global_init and ttm_mem_global_release, respectively.
	</para>
	<para>
	  Once your global TTM accounting structure is set up and initialized
	  by calling ttm_global_item_ref() on it,
	  you need to create a buffer object TTM to
	  provide a pool for buffer object allocation by clients and the
	  kernel itself.  The type of this object should be TTM_GLOBAL_TTM_BO,
	  and its size should be sizeof(struct ttm_bo_global).  Again,
	  driver-specific init and release functions may be provided,
	  likely eventually calling ttm_bo_global_init() and
	  ttm_bo_global_release(), respectively.  Also, like the previous
	  object, ttm_global_item_ref() is used to create an initial reference
	  count for the TTM, which will call your initialization function.
	</para>
      </sect3>
      <sect3>
	<title>GEM initialization</title>
	<para>
	  GEM is an alternative to TTM, designed specifically for UMA
	  devices.  It has simpler initialization and execution requirements
	  than TTM, but has no VRAM management capability.  Core GEM
	  is initialized by calling drm_mm_init() to create
	  a GTT DRM MM object, which provides an address space pool for
	  object allocation.  In a KMS configuration, the driver
	  needs to allocate and initialize a command ring buffer following
	  core GEM initialization.  A UMA device usually has what is called a
	  "stolen" memory region, which provides space for the initial
	  framebuffer and large, contiguous memory regions required by the
	  device.  This space is not typically managed by GEM, and it must
	  be initialized separately into its own DRM MM object.
	</para>
	<para>
	  Initialization is driver-specific. In the case of Intel
	  integrated graphics chips like 965GM, GEM initialization can
	  be done by calling the internal GEM init function,
	  i915_gem_do_init().  Since the 965GM is a UMA device
	  (i.e. it doesn't have dedicated VRAM), GEM manages
	  making regular RAM available for GPU operations.  Memory set
	  aside by the BIOS (called "stolen" memory by the i915
	  driver) is managed by the DRM memrange allocator; the
	  rest of the aperture is managed by GEM.
	  <programlisting>
	    /* Basic memrange allocator for stolen space (aka vram) */
	    drm_memrange_init(&amp;dev_priv->vram, 0, prealloc_size);
	    /* Let GEM Manage from end of prealloc space to end of aperture */
	    i915_gem_do_init(dev, prealloc_size, agp_size);
	  </programlisting>
<!--!Edrivers/char/drm/drm_memrange.c-->
	</para>
	<para>
	  Once the memory manager has been set up, we may allocate the
	  command buffer.  In the i915 case, this is also done with a
	  GEM function, i915_gem_init_ringbuffer().
	</para>
      </sect3>
    </sect2>

    <sect2>
      <title>Output configuration</title>
      <para>
	The final initialization task is output configuration.  This involves:
	<itemizedlist>
	  <listitem>
	    Finding and initializing the CRTCs, encoders, and connectors
	    for the device.
	  </listitem>
	  <listitem>
	    Creating an initial configuration.
	  </listitem>
	  <listitem>
	    Registering a framebuffer console driver.
	  </listitem>
	</itemizedlist>
      </para>
      <sect3>
	<title>Output discovery and initialization</title>
	<para>
	  Several core functions exist to create CRTCs, encoders, and
	  connectors, namely: drm_crtc_init(), drm_connector_init(), and
	  drm_encoder_init(), along with several "helper" functions to
	  perform common tasks.
	</para>
	<para>
	  Connectors should be registered with sysfs once they've been
	  detected and initialized, using the
	  drm_sysfs_connector_add() function.  Likewise, when they're
	  removed from the system, they should be destroyed with
	  drm_sysfs_connector_remove().
	</para>
	<programlisting>
<![CDATA[
void intel_crt_init(struct drm_device *dev)
{
	struct drm_connector *connector;
	struct intel_output *intel_output;

	intel_output = kzalloc(sizeof(struct intel_output), GFP_KERNEL);
	if (!intel_output)
		return;

	connector = &intel_output->base;
	drm_connector_init(dev, &intel_output->base,
			   &intel_crt_connector_funcs, DRM_MODE_CONNECTOR_VGA);

	drm_encoder_init(dev, &intel_output->enc, &intel_crt_enc_funcs,
			 DRM_MODE_ENCODER_DAC);

	drm_mode_connector_attach_encoder(&intel_output->base,
					  &intel_output->enc);

	/* Set up the DDC bus. */
	intel_output->ddc_bus = intel_i2c_create(dev, GPIOA, "CRTDDC_A");
	if (!intel_output->ddc_bus) {
		dev_printk(KERN_ERR, &dev->pdev->dev, "DDC bus registration "
			   "failed.\n");
		return;
	}

	intel_output->type = INTEL_OUTPUT_ANALOG;
	connector->interlace_allowed = 0;
	connector->doublescan_allowed = 0;

	drm_encoder_helper_add(&intel_output->enc, &intel_crt_helper_funcs);
	drm_connector_helper_add(connector, &intel_crt_connector_helper_funcs);

	drm_sysfs_connector_add(connector);
}
]]>
	</programlisting>
	<para>
	  In the example above (again, taken from the i915 driver), a
	  CRT connector and encoder combination is created.  A device-specific
	  i2c bus is also created for fetching EDID data and
	  performing monitor detection.  Once the process is complete,
	  the new connector is registered with sysfs to make its
	  properties available to applications.
	</para>
	<sect4>
	  <title>Helper functions and core functions</title>
	  <para>
	    Since many PC-class graphics devices have similar display output
	    designs, the DRM provides a set of helper functions to make
	    output management easier.  The core helper routines handle
	    encoder re-routing and the disabling of unused functions following
	    mode setting.  Using the helpers is optional, but recommended for
	    devices with PC-style architectures (i.e. a set of display planes
	    for feeding pixels to encoders which are in turn routed to
	    connectors).  Devices with more complex requirements needing
	    finer grained management may opt to use the core callbacks
	    directly.
	  </para>
	  <para>
	    [Insert typical diagram here.]  [Insert OMAP style config here.]
	  </para>
	</sect4>
	<para>
	  Each encoder object needs to provide:
	  <itemizedlist>
	    <listitem>
	      A DPMS (basically on/off) function.
	    </listitem>
	    <listitem>
	      A mode-fixup function (for converting requested modes into
	      native hardware timings).
	    </listitem>
	    <listitem>
	      Functions (prepare, set, and commit) for use by the core DRM
	      helper functions.
	    </listitem>
	  </itemizedlist>
	  Connector helpers need to provide functions (mode-fetch, validity,
	  and encoder-matching) for returning an ideal encoder for a given
	  connector.  The core connector functions include a DPMS callback,
	  save/restore routines (deprecated), detection, mode probing,
	  property handling, and cleanup functions.
	</para>
<!--!Edrivers/char/drm/drm_crtc.h-->
<!--!Edrivers/char/drm/drm_crtc.c-->
<!--!Edrivers/char/drm/drm_crtc_helper.c-->
      </sect3>
    </sect2>
  </sect1>

  <!-- Internals: vblank handling -->

  <sect1>
    <title>VBlank event handling</title>
    <para>
      The DRM core exposes two vertical blank related ioctls:
      <variablelist>
        <varlistentry>
          <term>DRM_IOCTL_WAIT_VBLANK</term>
          <listitem>
            <para>
              This takes a struct drm_wait_vblank structure as its argument,
              and it is used to block or request a signal when a specified
              vblank event occurs.
            </para>
          </listitem>
        </varlistentry>
        <varlistentry>
          <term>DRM_IOCTL_MODESET_CTL</term>
          <listitem>
            <para>
              This should be called by application level drivers before and
              after mode setting, since on many devices the vertical blank
              counter is reset at that time.  Internally, the DRM snapshots
              the last vblank count when the ioctl is called with the
              _DRM_PRE_MODESET command, so that the counter won't go backwards
              (which is dealt with when _DRM_POST_MODESET is used).
            </para>
          </listitem>
        </varlistentry>
      </variablelist>
<!--!Edrivers/char/drm/drm_irq.c-->
    </para>
    <para>
      To support the functions above, the DRM core provides several
      helper functions for tracking vertical blank counters, and
      requires drivers to provide several callbacks:
      get_vblank_counter(), enable_vblank() and disable_vblank().  The
      core uses get_vblank_counter() to keep the counter accurate
      across interrupt disable periods.  It should return the current
      vertical blank event count, which is often tracked in a device
      register.  The enable and disable vblank callbacks should enable
      and disable vertical blank interrupts, respectively.  In the
      absence of DRM clients waiting on vblank events, the core DRM
      code uses the disable_vblank() function to disable
      interrupts, which saves power.  They are re-enabled again when
      a client calls the vblank wait ioctl above.
    </para>
    <para>
      A device that doesn't provide a count register may simply use an
      internal atomic counter incremented on every vertical blank
      interrupt (and then treat the enable_vblank() and disable_vblank()
      callbacks as no-ops).
    </para>
  </sect1>

  <sect1>
    <title>Memory management</title>
    <para>
      The memory manager lies at the heart of many DRM operations; it
      is required to support advanced client features like OpenGL
      pbuffers.  The DRM currently contains two memory managers: TTM
      and GEM.
    </para>

    <sect2>
      <title>The Translation Table Manager (TTM)</title>
      <para>
	TTM was developed by Tungsten Graphics, primarily by Thomas
	Hellström, and is intended to be a flexible, high performance
	graphics memory manager.
      </para>
      <para>
	Drivers wishing to support TTM must fill out a drm_bo_driver
	structure.
      </para>
      <para>
	TTM design background and information belongs here.
      </para>
    </sect2>

    <sect2>
      <title>The Graphics Execution Manager (GEM)</title>
      <para>
	GEM is an Intel project, authored by Eric Anholt and Keith
	Packard.  It provides simpler interfaces than TTM, and is well
	suited for UMA devices.
      </para>
      <para>
	GEM-enabled drivers must provide gem_init_object() and
	gem_free_object() callbacks to support the core memory
	allocation routines.  They should also provide several driver-specific
	ioctls to support command execution, pinning, buffer
	read &amp; write, mapping, and domain ownership transfers.
      </para>
      <para>
	On a fundamental level, GEM involves several operations:
	<itemizedlist>
	  <listitem>Memory allocation and freeing</listitem>
	  <listitem>Command execution</listitem>
	  <listitem>Aperture management at command execution time</listitem>
	</itemizedlist>
	Buffer object allocation is relatively
	straightforward and largely provided by Linux's shmem layer, which
	provides memory to back each object.  When mapped into the GTT
	or used in a command buffer, the backing pages for an object are
	flushed to memory and marked write combined so as to be coherent
	with the GPU.  Likewise, if the CPU accesses an object after the GPU
	has finished rendering to the object, then the object must be made
	coherent with the CPU's view
	of memory, usually involving GPU cache flushing of various kinds.
	This core CPU&lt;-&gt;GPU coherency management is provided by a
	device-specific ioctl, which evaluates an object's current domain and
	performs any necessary flushing or synchronization to put the object
	into the desired coherency domain (note that the object may be busy,
	i.e. an active render target; in that case, setting the domain
	blocks the client and waits for rendering to complete before
	performing any necessary flushing operations).
      </para>
      <para>
	Perhaps the most important GEM function is providing a command
	execution interface to clients.  Client programs construct command
	buffers containing references to previously allocated memory objects,
	and then submit them to GEM.  At that point, GEM takes care to bind
	all the objects into the GTT, execute the buffer, and provide
	necessary synchronization between clients accessing the same buffers.
	This often involves evicting some objects from the GTT and re-binding
	others (a fairly expensive operation), and providing relocation
	support which hides fixed GTT offsets from clients.  Clients must
	take care not to submit command buffers that reference more objects
	than can fit in the GTT; otherwise, GEM will reject them and no rendering
	will occur.  Similarly, if several objects in the buffer require
	fence registers to be allocated for correct rendering (e.g. 2D blits
	on pre-965 chips), care must be taken not to require more fence
	registers than are available to the client.  Such resource management
	should be abstracted from the client in libdrm.
      </para>
    </sect2>

  </sect1>

  <!-- Output management -->
  <sect1>
    <title>Output management</title>
    <para>
      At the core of the DRM output management code is a set of
      structures representing CRTCs, encoders, and connectors.
    </para>
    <para>
      A CRTC is an abstraction representing a part of the chip that
      contains a pointer to a scanout buffer.  Therefore, the number
      of CRTCs available determines how many independent scanout
      buffers can be active at any given time.  The CRTC structure
      contains several fields to support this: a pointer to some video
      memory, a display mode, and an (x, y) offset into the video
      memory to support panning or configurations where one piece of
      video memory spans multiple CRTCs.
    </para>
    <para>
      An encoder takes pixel data from a CRTC and converts it to a
      format suitable for any attached connectors.  On some devices,
      it may be possible to have a CRTC send data to more than one
      encoder.  In that case, both encoders would receive data from
      the same scanout buffer, resulting in a "cloned" display
      configuration across the connectors attached to each encoder.
    </para>
    <para>
      A connector is the final destination for pixel data on a device,
      and usually connects directly to an external display device like
      a monitor or laptop panel.  A connector can only be attached to
      one encoder at a time.  The connector is also the structure
      where information about the attached display is kept, so it
      contains fields for display data, EDID data, DPMS &amp;
      connection status, and information about modes supported on the
      attached displays.
    </para>
<!--!Edrivers/char/drm/drm_crtc.c-->
  </sect1>

  <sect1>
    <title>Framebuffer management</title>
    <para>
      Clients need to provide a framebuffer object which provides a source
      of pixels for a CRTC to deliver to the encoder(s) and ultimately the
      connector(s). A framebuffer is fundamentally a driver-specific memory
      object, made into an opaque handle by the DRM's addfb() function.
      Once a framebuffer has been created this way, it may be passed to the
      KMS mode setting routines for use in a completed configuration.
    </para>
  </sect1>

  <sect1>
    <title>Command submission &amp; fencing</title>
    <para>
      This should cover a few device-specific command submission
      implementations.
    </para>
  </sect1>

  <sect1>
    <title>Suspend/resume</title>
    <para>
      The DRM core provides some suspend/resume code, but drivers
      wanting full suspend/resume support should provide save() and
      restore() functions.  These are called at suspend,
      hibernate, or resume time, and should perform any state save or
      restore required by your device across suspend or hibernate
      states.
    </para>
  </sect1>

  <sect1>
    <title>DMA services</title>
    <para>
      This should cover how DMA mapping etc. is supported by the core.
      These functions are deprecated and should not be used.
    </para>
  </sect1>
  </chapter>

  <!-- External interfaces -->

  <chapter id="drmExternals">
    <title>Userland interfaces</title>
    <para>
      The DRM core exports several interfaces to applications,
      generally intended to be used through corresponding libdrm
      wrapper functions.  In addition, drivers export device-specific
      interfaces for use by userspace drivers &amp; device-aware
      applications through ioctls and sysfs files.
    </para>
    <para>
      External interfaces include: memory mapping, context management,
      DMA operations, AGP management, vblank control, fence
      management, memory management, and output management.
    </para>
    <para>
      Cover generic ioctls and sysfs layout here.  We only need high-level
      info, since man pages should cover the rest.
    </para>
  </chapter>

  <!-- API reference -->

  <appendix id="drmDriverApi">
    <title>DRM Driver API</title>
    <para>
      Include auto-generated API reference here (need to reference it
      from paragraphs above too).
    </para>
  </appendix>

</book>