linux/Documentation/DocBook/drm.tmpl
<<
>>
Prefs
   1<?xml version="1.0" encoding="UTF-8"?>
   2<!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN"
   3        "http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []>
   4
   5<book id="drmDevelopersGuide">
   6  <bookinfo>
   7    <title>Linux DRM Developer's Guide</title>
   8
   9    <authorgroup>
  10      <author>
  11        <firstname>Jesse</firstname>
  12        <surname>Barnes</surname>
  13        <contrib>Initial version</contrib>
  14        <affiliation>
  15          <orgname>Intel Corporation</orgname>
  16          <address>
  17            <email>jesse.barnes@intel.com</email>
  18          </address>
  19        </affiliation>
  20      </author>
  21      <author>
  22        <firstname>Laurent</firstname>
  23        <surname>Pinchart</surname>
  24        <contrib>Driver internals</contrib>
  25        <affiliation>
  26          <orgname>Ideas on board SPRL</orgname>
  27          <address>
  28            <email>laurent.pinchart@ideasonboard.com</email>
  29          </address>
  30        </affiliation>
  31      </author>
  32    </authorgroup>
  33
  34    <copyright>
  35      <year>2008-2009</year>
  36      <year>2012</year>
  37      <holder>Intel Corporation</holder>
  38      <holder>Laurent Pinchart</holder>
  39    </copyright>
  40
  41    <legalnotice>
  42      <para>
  43        The contents of this file may be used under the terms of the GNU
  44        General Public License version 2 (the "GPL") as distributed in
  45        the kernel source COPYING file.
  46      </para>
  47    </legalnotice>
  48
  49    <revhistory>
  50      <!-- Put document revisions here, newest first. -->
  51      <revision>
  52        <revnumber>1.0</revnumber>
  53        <date>2012-07-13</date>
  54        <authorinitials>LP</authorinitials>
  55        <revremark>Added extensive documentation about driver internals.
  56        </revremark>
  57      </revision>
  58    </revhistory>
  59  </bookinfo>
  60
  61<toc></toc>
  62
  63  <!-- Introduction -->
  64
  65  <chapter id="drmIntroduction">
  66    <title>Introduction</title>
  67    <para>
  68      The Linux DRM layer contains code intended to support the needs
  69      of complex graphics devices, usually containing programmable
  70      pipelines well suited to 3D graphics acceleration.  Graphics
  71      drivers in the kernel may make use of DRM functions to make
  72      tasks like memory management, interrupt handling and DMA easier,
  73      and provide a uniform interface to applications.
  74    </para>
  75    <para>
  76      A note on versions: this guide covers features found in the DRM
  77      tree, including the TTM memory manager, output configuration and
  78      mode setting, and the new vblank internals, in addition to all
  79      the regular features found in current kernels.
  80    </para>
  81    <para>
  82      [Insert diagram of typical DRM stack here]
  83    </para>
  84  </chapter>
  85
  86  <!-- Internals -->
  87
  88  <chapter id="drmInternals">
  89    <title>DRM Internals</title>
  90    <para>
  91      This chapter documents DRM internals relevant to driver authors
  92      and developers working to add support for the latest features to
  93      existing drivers.
  94    </para>
  95    <para>
  96      First, we go over some typical driver initialization
  97      requirements, like setting up command buffers, creating an
  98      initial output configuration, and initializing core services.
  99      Subsequent sections cover core internals in more detail,
 100      providing implementation notes and examples.
 101    </para>
 102    <para>
 103      The DRM layer provides several services to graphics drivers,
 104      many of them driven by the application interfaces it provides
 105      through libdrm, the library that wraps most of the DRM ioctls.
 106      These include vblank event handling, memory
 107      management, output management, framebuffer management, command
 108      submission &amp; fencing, suspend/resume support, and DMA
 109      services.
 110    </para>
 111
 112  <!-- Internals: driver init -->
 113
 114  <sect1>
 115    <title>Driver Initialization</title>
 116    <para>
 117      At the core of every DRM driver is a <structname>drm_driver</structname>
 118      structure. Drivers typically statically initialize a drm_driver structure,
 119      and then pass it to one of the <function>drm_*_init()</function> functions
 120      to register it with the DRM subsystem.
 121    </para>
 122    <para>
 123      The <structname>drm_driver</structname> structure contains static
 124      information that describes the driver and features it supports, and
 125      pointers to methods that the DRM core will call to implement the DRM API.
 126      We will first go through the <structname>drm_driver</structname> static
 127      information fields, and will then describe individual operations in
 128      details as they get used in later sections.
 129    </para>
 130    <sect2>
 131      <title>Driver Information</title>
 132      <sect3>
 133        <title>Driver Features</title>
 134        <para>
 135          Drivers inform the DRM core about their requirements and supported
 136          features by setting appropriate flags in the
 137          <structfield>driver_features</structfield> field. Since those flags
 138          influence the DRM core behaviour since registration time, most of them
 139          must be set to registering the <structname>drm_driver</structname>
 140          instance.
 141        </para>
 142        <synopsis>u32 driver_features;</synopsis>
 143        <variablelist>
 144          <title>Driver Feature Flags</title>
 145          <varlistentry>
 146            <term>DRIVER_USE_AGP</term>
 147            <listitem><para>
 148              Driver uses AGP interface, the DRM core will manage AGP resources.
 149            </para></listitem>
 150          </varlistentry>
 151          <varlistentry>
 152            <term>DRIVER_REQUIRE_AGP</term>
 153            <listitem><para>
 154              Driver needs AGP interface to function. AGP initialization failure
 155              will become a fatal error.
 156            </para></listitem>
 157          </varlistentry>
 158          <varlistentry>
 159            <term>DRIVER_USE_MTRR</term>
 160            <listitem><para>
 161              Driver uses MTRR interface for mapping memory, the DRM core will
 162              manage MTRR resources. Deprecated.
 163            </para></listitem>
 164          </varlistentry>
 165          <varlistentry>
 166            <term>DRIVER_PCI_DMA</term>
 167            <listitem><para>
 168              Driver is capable of PCI DMA, mapping of PCI DMA buffers to
 169              userspace will be enabled. Deprecated.
 170            </para></listitem>
 171          </varlistentry>
 172          <varlistentry>
 173            <term>DRIVER_SG</term>
 174            <listitem><para>
 175              Driver can perform scatter/gather DMA, allocation and mapping of
 176              scatter/gather buffers will be enabled. Deprecated.
 177            </para></listitem>
 178          </varlistentry>
 179          <varlistentry>
 180            <term>DRIVER_HAVE_DMA</term>
 181            <listitem><para>
 182              Driver supports DMA, the userspace DMA API will be supported.
 183              Deprecated.
 184            </para></listitem>
 185          </varlistentry>
 186          <varlistentry>
 187            <term>DRIVER_HAVE_IRQ</term><term>DRIVER_IRQ_SHARED</term>
 188            <listitem><para>
 189              DRIVER_HAVE_IRQ indicates whether the driver has an IRQ handler. The
 190              DRM core will automatically register an interrupt handler when the
 191              flag is set. DRIVER_IRQ_SHARED indicates whether the device &amp;
 192              handler support shared IRQs (note that this is required of PCI
 193              drivers).
 194            </para></listitem>
 195          </varlistentry>
 196          <varlistentry>
 197            <term>DRIVER_IRQ_VBL</term>
 198            <listitem><para>Unused. Deprecated.</para></listitem>
 199          </varlistentry>
 200          <varlistentry>
 201            <term>DRIVER_DMA_QUEUE</term>
 202            <listitem><para>
 203              Should be set if the driver queues DMA requests and completes them
 204              asynchronously.  Deprecated.
 205            </para></listitem>
 206          </varlistentry>
 207          <varlistentry>
 208            <term>DRIVER_FB_DMA</term>
 209            <listitem><para>
 210              Driver supports DMA to/from the framebuffer, mapping of frambuffer
 211              DMA buffers to userspace will be supported. Deprecated.
 212            </para></listitem>
 213          </varlistentry>
 214          <varlistentry>
 215            <term>DRIVER_IRQ_VBL2</term>
 216            <listitem><para>Unused. Deprecated.</para></listitem>
 217          </varlistentry>
 218          <varlistentry>
 219            <term>DRIVER_GEM</term>
 220            <listitem><para>
 221              Driver use the GEM memory manager.
 222            </para></listitem>
 223          </varlistentry>
 224          <varlistentry>
 225            <term>DRIVER_MODESET</term>
 226            <listitem><para>
 227              Driver supports mode setting interfaces (KMS).
 228            </para></listitem>
 229          </varlistentry>
 230          <varlistentry>
 231            <term>DRIVER_PRIME</term>
 232            <listitem><para>
 233              Driver implements DRM PRIME buffer sharing.
 234            </para></listitem>
 235          </varlistentry>
 236        </variablelist>
 237      </sect3>
 238      <sect3>
 239        <title>Major, Minor and Patchlevel</title>
 240        <synopsis>int major;
 241int minor;
 242int patchlevel;</synopsis>
 243        <para>
 244          The DRM core identifies driver versions by a major, minor and patch
 245          level triplet. The information is printed to the kernel log at
 246          initialization time and passed to userspace through the
 247          DRM_IOCTL_VERSION ioctl.
 248        </para>
 249        <para>
 250          The major and minor numbers are also used to verify the requested driver
 251          API version passed to DRM_IOCTL_SET_VERSION. When the driver API changes
 252          between minor versions, applications can call DRM_IOCTL_SET_VERSION to
 253          select a specific version of the API. If the requested major isn't equal
 254          to the driver major, or the requested minor is larger than the driver
 255          minor, the DRM_IOCTL_SET_VERSION call will return an error. Otherwise
 256          the driver's set_version() method will be called with the requested
 257          version.
 258        </para>
 259      </sect3>
 260      <sect3>
 261        <title>Name, Description and Date</title>
 262        <synopsis>char *name;
 263char *desc;
 264char *date;</synopsis>
 265        <para>
 266          The driver name is printed to the kernel log at initialization time,
 267          used for IRQ registration and passed to userspace through
 268          DRM_IOCTL_VERSION.
 269        </para>
 270        <para>
 271          The driver description is a purely informative string passed to
 272          userspace through the DRM_IOCTL_VERSION ioctl and otherwise unused by
 273          the kernel.
 274        </para>
 275        <para>
 276          The driver date, formatted as YYYYMMDD, is meant to identify the date of
 277          the latest modification to the driver. However, as most drivers fail to
 278          update it, its value is mostly useless. The DRM core prints it to the
 279          kernel log at initialization time and passes it to userspace through the
 280          DRM_IOCTL_VERSION ioctl.
 281        </para>
 282      </sect3>
 283    </sect2>
 284    <sect2>
 285      <title>Driver Load</title>
 286      <para>
 287        The <methodname>load</methodname> method is the driver and device
 288        initialization entry point. The method is responsible for allocating and
 289        initializing driver private data, specifying supported performance
 290        counters, performing resource allocation and mapping (e.g. acquiring
 291        clocks, mapping registers or allocating command buffers), initializing
 292        the memory manager (<xref linkend="drm-memory-management"/>), installing
 293        the IRQ handler (<xref linkend="drm-irq-registration"/>), setting up
 294        vertical blanking handling (<xref linkend="drm-vertical-blank"/>), mode
 295        setting (<xref linkend="drm-mode-setting"/>) and initial output
 296        configuration (<xref linkend="drm-kms-init"/>).
 297      </para>
 298      <note><para>
 299        If compatibility is a concern (e.g. with drivers converted over from
 300        User Mode Setting to Kernel Mode Setting), care must be taken to prevent
 301        device initialization and control that is incompatible with currently
 302        active userspace drivers. For instance, if user level mode setting
 303        drivers are in use, it would be problematic to perform output discovery
 304        &amp; configuration at load time. Likewise, if user-level drivers
 305        unaware of memory management are in use, memory management and command
 306        buffer setup may need to be omitted. These requirements are
 307        driver-specific, and care needs to be taken to keep both old and new
 308        applications and libraries working.
 309      </para></note>
 310      <synopsis>int (*load) (struct drm_device *, unsigned long flags);</synopsis>
 311      <para>
 312        The method takes two arguments, a pointer to the newly created
 313        <structname>drm_device</structname> and flags. The flags are used to
 314        pass the <structfield>driver_data</structfield> field of the device id
 315        corresponding to the device passed to <function>drm_*_init()</function>.
 316        Only PCI devices currently use this, USB and platform DRM drivers have
 317        their <methodname>load</methodname> method called with flags to 0.
 318      </para>
 319      <sect3>
 320        <title>Driver Private &amp; Performance Counters</title>
 321        <para>
 322          The driver private hangs off the main
 323          <structname>drm_device</structname> structure and can be used for
 324          tracking various device-specific bits of information, like register
 325          offsets, command buffer status, register state for suspend/resume, etc.
 326          At load time, a driver may simply allocate one and set
 327          <structname>drm_device</structname>.<structfield>dev_priv</structfield>
 328          appropriately; it should be freed and
 329          <structname>drm_device</structname>.<structfield>dev_priv</structfield>
 330          set to NULL when the driver is unloaded.
 331        </para>
 332        <para>
 333          DRM supports several counters which were used for rough performance
 334          characterization. This stat counter system is deprecated and should not
 335          be used. If performance monitoring is desired, the developer should
 336          investigate and potentially enhance the kernel perf and tracing
 337          infrastructure to export GPU related performance information for
 338          consumption by performance monitoring tools and applications.
 339        </para>
 340      </sect3>
 341      <sect3 id="drm-irq-registration">
 342        <title>IRQ Registration</title>
 343        <para>
 344          The DRM core tries to facilitate IRQ handler registration and
 345          unregistration by providing <function>drm_irq_install</function> and
 346          <function>drm_irq_uninstall</function> functions. Those functions only
 347          support a single interrupt per device.
 348        </para>
 349  <!--!Fdrivers/char/drm/drm_irq.c drm_irq_install-->
 350        <para>
 351          Both functions get the device IRQ by calling
 352          <function>drm_dev_to_irq</function>. This inline function will call a
 353          bus-specific operation to retrieve the IRQ number. For platform devices,
 354          <function>platform_get_irq</function>(..., 0) is used to retrieve the
 355          IRQ number.
 356        </para>
 357        <para>
 358          <function>drm_irq_install</function> starts by calling the
 359          <methodname>irq_preinstall</methodname> driver operation. The operation
 360          is optional and must make sure that the interrupt will not get fired by
 361          clearing all pending interrupt flags or disabling the interrupt.
 362        </para>
 363        <para>
 364          The IRQ will then be requested by a call to
 365          <function>request_irq</function>. If the DRIVER_IRQ_SHARED driver
 366          feature flag is set, a shared (IRQF_SHARED) IRQ handler will be
 367          requested.
 368        </para>
 369        <para>
 370          The IRQ handler function must be provided as the mandatory irq_handler
 371          driver operation. It will get passed directly to
 372          <function>request_irq</function> and thus has the same prototype as all
 373          IRQ handlers. It will get called with a pointer to the DRM device as the
 374          second argument.
 375        </para>
 376        <para>
 377          Finally the function calls the optional
 378          <methodname>irq_postinstall</methodname> driver operation. The operation
 379          usually enables interrupts (excluding the vblank interrupt, which is
 380          enabled separately), but drivers may choose to enable/disable interrupts
 381          at a different time.
 382        </para>
 383        <para>
 384          <function>drm_irq_uninstall</function> is similarly used to uninstall an
 385          IRQ handler. It starts by waking up all processes waiting on a vblank
 386          interrupt to make sure they don't hang, and then calls the optional
 387          <methodname>irq_uninstall</methodname> driver operation. The operation
 388          must disable all hardware interrupts. Finally the function frees the IRQ
 389          by calling <function>free_irq</function>.
 390        </para>
 391      </sect3>
 392      <sect3>
 393        <title>Memory Manager Initialization</title>
 394        <para>
 395          Every DRM driver requires a memory manager which must be initialized at
 396          load time. DRM currently contains two memory managers, the Translation
 397          Table Manager (TTM) and the Graphics Execution Manager (GEM).
 398          This document describes the use of the GEM memory manager only. See
 399          <xref linkend="drm-memory-management"/> for details.
 400        </para>
 401      </sect3>
 402      <sect3>
 403        <title>Miscellaneous Device Configuration</title>
 404        <para>
 405          Another task that may be necessary for PCI devices during configuration
 406          is mapping the video BIOS. On many devices, the VBIOS describes device
 407          configuration, LCD panel timings (if any), and contains flags indicating
 408          device state. Mapping the BIOS can be done using the pci_map_rom() call,
 409          a convenience function that takes care of mapping the actual ROM,
 410          whether it has been shadowed into memory (typically at address 0xc0000)
 411          or exists on the PCI device in the ROM BAR. Note that after the ROM has
 412          been mapped and any necessary information has been extracted, it should
 413          be unmapped; on many devices, the ROM address decoder is shared with
 414          other BARs, so leaving it mapped could cause undesired behaviour like
 415          hangs or memory corruption.
 416  <!--!Fdrivers/pci/rom.c pci_map_rom-->
 417        </para>
 418      </sect3>
 419    </sect2>
 420  </sect1>
 421
 422  <!-- Internals: memory management -->
 423
 424  <sect1 id="drm-memory-management">
 425    <title>Memory management</title>
 426    <para>
 427      Modern Linux systems require large amount of graphics memory to store
 428      frame buffers, textures, vertices and other graphics-related data. Given
 429      the very dynamic nature of many of that data, managing graphics memory
 430      efficiently is thus crucial for the graphics stack and plays a central
 431      role in the DRM infrastructure.
 432    </para>
 433    <para>
 434      The DRM core includes two memory managers, namely Translation Table Maps
 435      (TTM) and Graphics Execution Manager (GEM). TTM was the first DRM memory
 436      manager to be developed and tried to be a one-size-fits-them all
 437      solution. It provides a single userspace API to accomodate the need of
 438      all hardware, supporting both Unified Memory Architecture (UMA) devices
 439      and devices with dedicated video RAM (i.e. most discrete video cards).
 440      This resulted in a large, complex piece of code that turned out to be
 441      hard to use for driver development.
 442    </para>
 443    <para>
 444      GEM started as an Intel-sponsored project in reaction to TTM's
 445      complexity. Its design philosophy is completely different: instead of
 446      providing a solution to every graphics memory-related problems, GEM
 447      identified common code between drivers and created a support library to
 448      share it. GEM has simpler initialization and execution requirements than
 449      TTM, but has no video RAM management capabitilies and is thus limited to
 450      UMA devices.
 451    </para>
 452    <sect2>
 453      <title>The Translation Table Manager (TTM)</title>
 454      <para>
 455        TTM design background and information belongs here.
 456      </para>
 457      <sect3>
 458        <title>TTM initialization</title>
 459        <warning><para>This section is outdated.</para></warning>
 460        <para>
 461          Drivers wishing to support TTM must fill out a drm_bo_driver
 462          structure. The structure contains several fields with function
 463          pointers for initializing the TTM, allocating and freeing memory,
 464          waiting for command completion and fence synchronization, and memory
 465          migration. See the radeon_ttm.c file for an example of usage.
 466        </para>
 467        <para>
 468          The ttm_global_reference structure is made up of several fields:
 469        </para>
 470        <programlisting>
 471          struct ttm_global_reference {
 472                enum ttm_global_types global_type;
 473                size_t size;
 474                void *object;
 475                int (*init) (struct ttm_global_reference *);
 476                void (*release) (struct ttm_global_reference *);
 477          };
 478        </programlisting>
 479        <para>
 480          There should be one global reference structure for your memory
 481          manager as a whole, and there will be others for each object
 482          created by the memory manager at runtime.  Your global TTM should
 483          have a type of TTM_GLOBAL_TTM_MEM.  The size field for the global
 484          object should be sizeof(struct ttm_mem_global), and the init and
 485          release hooks should point at your driver-specific init and
 486          release routines, which probably eventually call
 487          ttm_mem_global_init and ttm_mem_global_release, respectively.
 488        </para>
 489        <para>
 490          Once your global TTM accounting structure is set up and initialized
 491          by calling ttm_global_item_ref() on it,
 492          you need to create a buffer object TTM to
 493          provide a pool for buffer object allocation by clients and the
 494          kernel itself.  The type of this object should be TTM_GLOBAL_TTM_BO,
 495          and its size should be sizeof(struct ttm_bo_global).  Again,
 496          driver-specific init and release functions may be provided,
 497          likely eventually calling ttm_bo_global_init() and
 498          ttm_bo_global_release(), respectively.  Also, like the previous
 499          object, ttm_global_item_ref() is used to create an initial reference
 500          count for the TTM, which will call your initialization function.
 501        </para>
 502      </sect3>
 503    </sect2>
 504    <sect2 id="drm-gem">
 505      <title>The Graphics Execution Manager (GEM)</title>
 506      <para>
 507        The GEM design approach has resulted in a memory manager that doesn't
 508        provide full coverage of all (or even all common) use cases in its
 509        userspace or kernel API. GEM exposes a set of standard memory-related
 510        operations to userspace and a set of helper functions to drivers, and let
 511        drivers implement hardware-specific operations with their own private API.
 512      </para>
 513      <para>
 514        The GEM userspace API is described in the
 515        <ulink url="http://lwn.net/Articles/283798/"><citetitle>GEM - the Graphics
 516        Execution Manager</citetitle></ulink> article on LWN. While slightly
 517        outdated, the document provides a good overview of the GEM API principles.
 518        Buffer allocation and read and write operations, described as part of the
 519        common GEM API, are currently implemented using driver-specific ioctls.
 520      </para>
 521      <para>
 522        GEM is data-agnostic. It manages abstract buffer objects without knowing
 523        what individual buffers contain. APIs that require knowledge of buffer
 524        contents or purpose, such as buffer allocation or synchronization
 525        primitives, are thus outside of the scope of GEM and must be implemented
 526        using driver-specific ioctls.
 527      </para>
 528      <para>
 529        On a fundamental level, GEM involves several operations:
 530        <itemizedlist>
 531          <listitem>Memory allocation and freeing</listitem>
 532          <listitem>Command execution</listitem>
 533          <listitem>Aperture management at command execution time</listitem>
 534        </itemizedlist>
 535        Buffer object allocation is relatively straightforward and largely
 536        provided by Linux's shmem layer, which provides memory to back each
 537        object.
 538      </para>
 539      <para>
 540        Device-specific operations, such as command execution, pinning, buffer
 541        read &amp; write, mapping, and domain ownership transfers are left to
 542        driver-specific ioctls.
 543      </para>
 544      <sect3>
 545        <title>GEM Initialization</title>
 546        <para>
 547          Drivers that use GEM must set the DRIVER_GEM bit in the struct
 548          <structname>drm_driver</structname>
 549          <structfield>driver_features</structfield> field. The DRM core will
 550          then automatically initialize the GEM core before calling the
 551          <methodname>load</methodname> operation. Behind the scene, this will
 552          create a DRM Memory Manager object which provides an address space
 553          pool for object allocation.
 554        </para>
 555        <para>
 556          In a KMS configuration, drivers need to allocate and initialize a
 557          command ring buffer following core GEM initialization if required by
 558          the hardware. UMA devices usually have what is called a "stolen"
 559          memory region, which provides space for the initial framebuffer and
 560          large, contiguous memory regions required by the device. This space is
 561          typically not managed by GEM, and must be initialized separately into
 562          its own DRM MM object.
 563        </para>
 564      </sect3>
 565      <sect3>
 566        <title>GEM Objects Creation</title>
 567        <para>
 568          GEM splits creation of GEM objects and allocation of the memory that
 569          backs them in two distinct operations.
 570        </para>
 571        <para>
 572          GEM objects are represented by an instance of struct
 573          <structname>drm_gem_object</structname>. Drivers usually need to extend
 574          GEM objects with private information and thus create a driver-specific
 575          GEM object structure type that embeds an instance of struct
 576          <structname>drm_gem_object</structname>.
 577        </para>
 578        <para>
 579          To create a GEM object, a driver allocates memory for an instance of its
 580          specific GEM object type and initializes the embedded struct
 581          <structname>drm_gem_object</structname> with a call to
 582          <function>drm_gem_object_init</function>. The function takes a pointer to
 583          the DRM device, a pointer to the GEM object and the buffer object size
 584          in bytes.
 585        </para>
 586        <para>
 587          GEM uses shmem to allocate anonymous pageable memory.
 588          <function>drm_gem_object_init</function> will create an shmfs file of
 589          the requested size and store it into the struct
 590          <structname>drm_gem_object</structname> <structfield>filp</structfield>
 591          field. The memory is used as either main storage for the object when the
 592          graphics hardware uses system memory directly or as a backing store
 593          otherwise.
 594        </para>
 595        <para>
 596          Drivers are responsible for the actual physical pages allocation by
 597          calling <function>shmem_read_mapping_page_gfp</function> for each page.
 598          Note that they can decide to allocate pages when initializing the GEM
 599          object, or to delay allocation until the memory is needed (for instance
 600          when a page fault occurs as a result of a userspace memory access or
 601          when the driver needs to start a DMA transfer involving the memory).
 602        </para>
 603        <para>
 604          Anonymous pageable memory allocation is not always desired, for instance
 605          when the hardware requires physically contiguous system memory as is
 606          often the case in embedded devices. Drivers can create GEM objects with
 607          no shmfs backing (called private GEM objects) by initializing them with
 608          a call to <function>drm_gem_private_object_init</function> instead of
 609          <function>drm_gem_object_init</function>. Storage for private GEM
 610          objects must be managed by drivers.
 611        </para>
 612        <para>
 613          Drivers that do not need to extend GEM objects with private information
 614          can call the <function>drm_gem_object_alloc</function> function to
 615          allocate and initialize a struct <structname>drm_gem_object</structname>
 616          instance. The GEM core will call the optional driver
 617          <methodname>gem_init_object</methodname> operation after initializing
 618          the GEM object with <function>drm_gem_object_init</function>.
 619          <synopsis>int (*gem_init_object) (struct drm_gem_object *obj);</synopsis>
 620        </para>
 621        <para>
 622          No alloc-and-init function exists for private GEM objects.
 623        </para>
 624      </sect3>
 625      <sect3>
 626        <title>GEM Objects Lifetime</title>
 627        <para>
 628          All GEM objects are reference-counted by the GEM core. References can be
 629          acquired and release by <function>calling drm_gem_object_reference</function>
 630          and <function>drm_gem_object_unreference</function> respectively. The
 631          caller must hold the <structname>drm_device</structname>
 632          <structfield>struct_mutex</structfield> lock. As a convenience, GEM
 633          provides the <function>drm_gem_object_reference_unlocked</function> and
 634          <function>drm_gem_object_unreference_unlocked</function> functions that
 635          can be called without holding the lock.
 636        </para>
 637        <para>
 638          When the last reference to a GEM object is released the GEM core calls
 639          the <structname>drm_driver</structname>
 640          <methodname>gem_free_object</methodname> operation. That operation is
 641          mandatory for GEM-enabled drivers and must free the GEM object and all
 642          associated resources.
 643        </para>
 644        <para>
 645          <synopsis>void (*gem_free_object) (struct drm_gem_object *obj);</synopsis>
 646          Drivers are responsible for freeing all GEM object resources, including
 647          the resources created by the GEM core. If an mmap offset has been
 648          created for the object (in which case
 649          <structname>drm_gem_object</structname>::<structfield>map_list</structfield>::<structfield>map</structfield>
 650          is not NULL) it must be freed by a call to
 651          <function>drm_gem_free_mmap_offset</function>. The shmfs backing store
 652          must be released by calling <function>drm_gem_object_release</function>
 653          (that function can safely be called if no shmfs backing store has been
 654          created).
 655        </para>
 656      </sect3>
 657      <sect3>
 658        <title>GEM Objects Naming</title>
 659        <para>
 660          Communication between userspace and the kernel refers to GEM objects
 661          using local handles, global names or, more recently, file descriptors.
 662          All of those are 32-bit integer values; the usual Linux kernel limits
 663          apply to the file descriptors.
 664        </para>
 665        <para>
 666          GEM handles are local to a DRM file. Applications get a handle to a GEM
 667          object through a driver-specific ioctl, and can use that handle to refer
 668          to the GEM object in other standard or driver-specific ioctls. Closing a
 669          DRM file handle frees all its GEM handles and dereferences the
 670          associated GEM objects.
 671        </para>
 672        <para>
 673          To create a handle for a GEM object drivers call
 674          <function>drm_gem_handle_create</function>. The function takes a pointer
 675          to the DRM file and the GEM object and returns a locally unique handle.
 676          When the handle is no longer needed drivers delete it with a call to
 677          <function>drm_gem_handle_delete</function>. Finally the GEM object
 678          associated with a handle can be retrieved by a call to
 679          <function>drm_gem_object_lookup</function>.
 680        </para>
 681        <para>
 682          Handles don't take ownership of GEM objects, they only take a reference
 683          to the object that will be dropped when the handle is destroyed. To
 684          avoid leaking GEM objects, drivers must make sure they drop the
 685          reference(s) they own (such as the initial reference taken at object
 686          creation time) as appropriate, without any special consideration for the
 687          handle. For example, in the particular case of combined GEM object and
 688          handle creation in the implementation of the
 689          <methodname>dumb_create</methodname> operation, drivers must drop the
 690          initial reference to the GEM object before returning the handle.
 691        </para>
 692        <para>
 693          GEM names are similar in purpose to handles but are not local to DRM
 694          files. They can be passed between processes to reference a GEM object
 695          globally. Names can't be used directly to refer to objects in the DRM
 696          API, applications must convert handles to names and names to handles
 697          using the DRM_IOCTL_GEM_FLINK and DRM_IOCTL_GEM_OPEN ioctls
 698          respectively. The conversion is handled by the DRM core without any
 699          driver-specific support.
 700        </para>
 701        <para>
 702          Similar to global names, GEM file descriptors are also used to share GEM
 703          objects across processes. They offer additional security: as file
 704          descriptors must be explictly sent over UNIX domain sockets to be shared
 705          between applications, they can't be guessed like the globally unique GEM
 706          names.
 707        </para>
 708        <para>
 709          Drivers that support GEM file descriptors, also known as the DRM PRIME
 710          API, must set the DRIVER_PRIME bit in the struct
 711          <structname>drm_driver</structname>
 712          <structfield>driver_features</structfield> field, and implement the
 713          <methodname>prime_handle_to_fd</methodname> and
 714          <methodname>prime_fd_to_handle</methodname> operations.
 715        </para>
 716        <para>
 717          <synopsis>int (*prime_handle_to_fd)(struct drm_device *dev,
 718                            struct drm_file *file_priv, uint32_t handle,
 719                            uint32_t flags, int *prime_fd);
 720  int (*prime_fd_to_handle)(struct drm_device *dev,
 721                            struct drm_file *file_priv, int prime_fd,
 722                            uint32_t *handle);</synopsis>
 723          Those two operations convert a handle to a PRIME file descriptor and
 724          vice versa. Drivers must use the kernel dma-buf buffer sharing framework
 725          to manage the PRIME file descriptors.
 726        </para>
 727        <para>
 728          While non-GEM drivers must implement the operations themselves, GEM
 729          drivers must use the <function>drm_gem_prime_handle_to_fd</function>
 730          and <function>drm_gem_prime_fd_to_handle</function> helper functions.
 731          Those helpers rely on the driver
 732          <methodname>gem_prime_export</methodname> and
 733          <methodname>gem_prime_import</methodname> operations to create a dma-buf
 734          instance from a GEM object (dma-buf exporter role) and to create a GEM
 735          object from a dma-buf instance (dma-buf importer role).
 736        </para>
 737        <para>
 738          <synopsis>struct dma_buf * (*gem_prime_export)(struct drm_device *dev,
 739                                       struct drm_gem_object *obj,
 740                                       int flags);
 741  struct drm_gem_object * (*gem_prime_import)(struct drm_device *dev,
 742                                              struct dma_buf *dma_buf);</synopsis>
 743          These two operations are mandatory for GEM drivers that support DRM
 744          PRIME.
 745        </para>
 746      </sect3>
 747      <sect3 id="drm-gem-objects-mapping">
 748        <title>GEM Objects Mapping</title>
 749        <para>
 750          Because mapping operations are fairly heavyweight GEM favours
 751          read/write-like access to buffers, implemented through driver-specific
 752          ioctls, over mapping buffers to userspace. However, when random access
 753          to the buffer is needed (to perform software rendering for instance),
 754          direct access to the object can be more efficient.
 755        </para>
 756        <para>
 757          The mmap system call can't be used directly to map GEM objects, as they
 758          don't have their own file handle. Two alternative methods currently
 759          co-exist to map GEM objects to userspace. The first method uses a
 760          driver-specific ioctl to perform the mapping operation, calling
 761          <function>do_mmap</function> under the hood. This is often considered
 762          dubious, seems to be discouraged for new GEM-enabled drivers, and will
 763          thus not be described here.
 764        </para>
 765        <para>
 766          The second method uses the mmap system call on the DRM file handle.
 767          <synopsis>void *mmap(void *addr, size_t length, int prot, int flags, int fd,
 768             off_t offset);</synopsis>
 769          DRM identifies the GEM object to be mapped by a fake offset passed
 770          through the mmap offset argument. Prior to being mapped, a GEM object
 771          must thus be associated with a fake offset. To do so, drivers must call
 772          <function>drm_gem_create_mmap_offset</function> on the object. The
 773          function allocates a fake offset range from a pool and stores the
 774          offset divided by PAGE_SIZE in
 775          <literal>obj-&gt;map_list.hash.key</literal>. Care must be taken not to
 776          call <function>drm_gem_create_mmap_offset</function> if a fake offset
 777          has already been allocated for the object. This can be tested by
 778          <literal>obj-&gt;map_list.map</literal> being non-NULL.
 779        </para>
 780        <para>
 781          Once allocated, the fake offset value
 782          (<literal>obj-&gt;map_list.hash.key &lt;&lt; PAGE_SHIFT</literal>)
 783          must be passed to the application in a driver-specific way and can then
 784          be used as the mmap offset argument.
 785        </para>
 786        <para>
 787          The GEM core provides a helper method <function>drm_gem_mmap</function>
 788          to handle object mapping. The method can be set directly as the mmap
 789          file operation handler. It will look up the GEM object based on the
 790          offset value and set the VMA operations to the
 791          <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
 792          field. Note that <function>drm_gem_mmap</function> doesn't map memory to
 793          userspace, but relies on the driver-provided fault handler to map pages
 794          individually.
 795        </para>
 796        <para>
 797          To use <function>drm_gem_mmap</function>, drivers must fill the struct
 798          <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
 799          field with a pointer to VM operations.
 800        </para>
 801        <para>
 802          <synopsis>struct vm_operations_struct *gem_vm_ops
 803
 804  struct vm_operations_struct {
 805          void (*open)(struct vm_area_struct * area);
 806          void (*close)(struct vm_area_struct * area);
 807          int (*fault)(struct vm_area_struct *vma, struct vm_fault *vmf);
 808  };</synopsis>
 809        </para>
 810        <para>
 811          The <methodname>open</methodname> and <methodname>close</methodname>
 812          operations must update the GEM object reference count. Drivers can use
 813          the <function>drm_gem_vm_open</function> and
 814          <function>drm_gem_vm_close</function> helper functions directly as open
 815          and close handlers.
 816        </para>
 817        <para>
 818          The fault operation handler is responsible for mapping individual pages
 819          to userspace when a page fault occurs. Depending on the memory
 820          allocation scheme, drivers can allocate pages at fault time, or can
 821          decide to allocate memory for the GEM object at the time the object is
 822          created.
 823        </para>
 824        <para>
 825          Drivers that want to map the GEM object upfront instead of handling page
 826          faults can implement their own mmap file operation handler.
 827        </para>
 828      </sect3>
 829      <sect3>
 830        <title>Dumb GEM Objects</title>
 831        <para>
 832          The GEM API doesn't standardize GEM objects creation and leaves it to
 833          driver-specific ioctls. While not an issue for full-fledged graphics
 834          stacks that include device-specific userspace components (in libdrm for
 835          instance), this limit makes DRM-based early boot graphics unnecessarily
 836          complex.
 837        </para>
 838        <para>
 839          Dumb GEM objects partly alleviate the problem by providing a standard
 840          API to create dumb buffers suitable for scanout, which can then be used
 841          to create KMS frame buffers.
 842        </para>
 843        <para>
 844          To support dumb GEM objects drivers must implement the
 845          <methodname>dumb_create</methodname>,
 846          <methodname>dumb_destroy</methodname> and
 847          <methodname>dumb_map_offset</methodname> operations.
 848        </para>
 849        <itemizedlist>
 850          <listitem>
 851            <synopsis>int (*dumb_create)(struct drm_file *file_priv, struct drm_device *dev,
 852                     struct drm_mode_create_dumb *args);</synopsis>
 853            <para>
 854              The <methodname>dumb_create</methodname> operation creates a GEM
 855              object suitable for scanout based on the width, height and depth
 856              from the struct <structname>drm_mode_create_dumb</structname>
 857              argument. It fills the argument's <structfield>handle</structfield>,
 858              <structfield>pitch</structfield> and <structfield>size</structfield>
 859              fields with a handle for the newly created GEM object and its line
 860              pitch and size in bytes.
 861            </para>
 862          </listitem>
 863          <listitem>
 864            <synopsis>int (*dumb_destroy)(struct drm_file *file_priv, struct drm_device *dev,
 865                      uint32_t handle);</synopsis>
 866            <para>
 867              The <methodname>dumb_destroy</methodname> operation destroys a dumb
 868              GEM object created by <methodname>dumb_create</methodname>.
 869            </para>
 870          </listitem>
 871          <listitem>
 872            <synopsis>int (*dumb_map_offset)(struct drm_file *file_priv, struct drm_device *dev,
 873                         uint32_t handle, uint64_t *offset);</synopsis>
 874            <para>
 875              The <methodname>dumb_map_offset</methodname> operation associates an
 876              mmap fake offset with the GEM object given by the handle and returns
 877              it. Drivers must use the
 878              <function>drm_gem_create_mmap_offset</function> function to
 879              associate the fake offset as described in
 880              <xref linkend="drm-gem-objects-mapping"/>.
 881            </para>
 882          </listitem>
 883        </itemizedlist>
 884      </sect3>
 885      <sect3>
 886        <title>Memory Coherency</title>
 887        <para>
 888          When mapped to the device or used in a command buffer, backing pages
 889          for an object are flushed to memory and marked write combined so as to
 890          be coherent with the GPU. Likewise, if the CPU accesses an object
 891          after the GPU has finished rendering to the object, then the object
 892          must be made coherent with the CPU's view of memory, usually involving
 893          GPU cache flushing of various kinds. This core CPU&lt;-&gt;GPU
 894          coherency management is provided by a device-specific ioctl, which
 895          evaluates an object's current domain and performs any necessary
 896          flushing or synchronization to put the object into the desired
 897          coherency domain (note that the object may be busy, i.e. an active
 898          render target; in that case, setting the domain blocks the client and
 899          waits for rendering to complete before performing any necessary
 900          flushing operations).
 901        </para>
 902      </sect3>
 903      <sect3>
 904        <title>Command Execution</title>
 905        <para>
 906          Perhaps the most important GEM function for GPU devices is providing a
 907          command execution interface to clients. Client programs construct
 908          command buffers containing references to previously allocated memory
 909          objects, and then submit them to GEM. At that point, GEM takes care to
 910          bind all the objects into the GTT, execute the buffer, and provide
 911          necessary synchronization between clients accessing the same buffers.
 912          This often involves evicting some objects from the GTT and re-binding
 913          others (a fairly expensive operation), and providing relocation
 914          support which hides fixed GTT offsets from clients. Clients must take
 915          care not to submit command buffers that reference more objects than
 916          can fit in the GTT; otherwise, GEM will reject them and no rendering
 917          will occur. Similarly, if several objects in the buffer require fence
 918          registers to be allocated for correct rendering (e.g. 2D blits on
 919          pre-965 chips), care must be taken not to require more fence registers
 920          than are available to the client. Such resource management should be
 921          abstracted from the client in libdrm.
 922        </para>
 923      </sect3>
 924    </sect2>
 925  </sect1>
 926
 927  <!-- Internals: mode setting -->
 928
 929  <sect1 id="drm-mode-setting">
 930    <title>Mode Setting</title>
 931    <para>
 932      Drivers must initialize the mode setting core by calling
 933      <function>drm_mode_config_init</function> on the DRM device. The function
 934      initializes the <structname>drm_device</structname>
 935      <structfield>mode_config</structfield> field and never fails. Once done,
 936      mode configuration must be setup by initializing the following fields.
 937    </para>
 938    <itemizedlist>
 939      <listitem>
 940        <synopsis>int min_width, min_height;
 941int max_width, max_height;</synopsis>
 942        <para>
 943          Minimum and maximum width and height of the frame buffers in pixel
 944          units.
 945        </para>
 946      </listitem>
 947      <listitem>
 948        <synopsis>struct drm_mode_config_funcs *funcs;</synopsis>
 949        <para>Mode setting functions.</para>
 950      </listitem>
 951    </itemizedlist>
 952    <sect2>
 953      <title>Frame Buffer Creation</title>
 954      <synopsis>struct drm_framebuffer *(*fb_create)(struct drm_device *dev,
 955                                     struct drm_file *file_priv,
 956                                     struct drm_mode_fb_cmd2 *mode_cmd);</synopsis>
 957      <para>
 958        Frame buffers are abstract memory objects that provide a source of
 959        pixels to scanout to a CRTC. Applications explicitly request the
 960        creation of frame buffers through the DRM_IOCTL_MODE_ADDFB(2) ioctls and
 961        receive an opaque handle that can be passed to the KMS CRTC control,
 962        plane configuration and page flip functions.
 963      </para>
 964      <para>
 965        Frame buffers rely on the underneath memory manager for low-level memory
 966        operations. When creating a frame buffer applications pass a memory
 967        handle (or a list of memory handles for multi-planar formats) through
 968        the <parameter>drm_mode_fb_cmd2</parameter> argument. This document
 969        assumes that the driver uses GEM, those handles thus reference GEM
 970        objects.
 971      </para>
 972      <para>
 973        Drivers must first validate the requested frame buffer parameters passed
 974        through the mode_cmd argument. In particular this is where invalid
 975        sizes, pixel formats or pitches can be caught.
 976      </para>
 977      <para>
 978        If the parameters are deemed valid, drivers then create, initialize and
 979        return an instance of struct <structname>drm_framebuffer</structname>.
 980        If desired the instance can be embedded in a larger driver-specific
 981        structure. The new instance is initialized with a call to
 982        <function>drm_framebuffer_init</function> which takes a pointer to DRM
 983        frame buffer operations (struct
 984        <structname>drm_framebuffer_funcs</structname>). Frame buffer operations are
 985        <itemizedlist>
 986          <listitem>
 987            <synopsis>int (*create_handle)(struct drm_framebuffer *fb,
 988                     struct drm_file *file_priv, unsigned int *handle);</synopsis>
 989            <para>
 990              Create a handle to the frame buffer underlying memory object. If
 991              the frame buffer uses a multi-plane format, the handle will
 992              reference the memory object associated with the first plane.
 993            </para>
 994            <para>
 995              Drivers call <function>drm_gem_handle_create</function> to create
 996              the handle.
 997            </para>
 998          </listitem>
 999          <listitem>
1000            <synopsis>void (*destroy)(struct drm_framebuffer *framebuffer);</synopsis>
1001            <para>
1002              Destroy the frame buffer object and frees all associated
1003              resources. Drivers must call
1004              <function>drm_framebuffer_cleanup</function> to free resources
1005              allocated by the DRM core for the frame buffer object, and must
1006              make sure to unreference all memory objects associated with the
1007              frame buffer. Handles created by the
1008              <methodname>create_handle</methodname> operation are released by
1009              the DRM core.
1010            </para>
1011          </listitem>
1012          <listitem>
1013            <synopsis>int (*dirty)(struct drm_framebuffer *framebuffer,
1014             struct drm_file *file_priv, unsigned flags, unsigned color,
1015             struct drm_clip_rect *clips, unsigned num_clips);</synopsis>
1016            <para>
1017              This optional operation notifies the driver that a region of the
1018              frame buffer has changed in response to a DRM_IOCTL_MODE_DIRTYFB
1019              ioctl call.
1020            </para>
1021          </listitem>
1022        </itemizedlist>
1023      </para>
1024      <para>
1025        After initializing the <structname>drm_framebuffer</structname>
1026        instance drivers must fill its <structfield>width</structfield>,
1027        <structfield>height</structfield>, <structfield>pitches</structfield>,
1028        <structfield>offsets</structfield>, <structfield>depth</structfield>,
1029        <structfield>bits_per_pixel</structfield> and
1030        <structfield>pixel_format</structfield> fields from the values passed
1031        through the <parameter>drm_mode_fb_cmd2</parameter> argument. They
1032        should call the <function>drm_helper_mode_fill_fb_struct</function>
1033        helper function to do so.
1034      </para>
1035    </sect2>
1036    <sect2>
1037      <title>Output Polling</title>
1038      <synopsis>void (*output_poll_changed)(struct drm_device *dev);</synopsis>
1039      <para>
1040        This operation notifies the driver that the status of one or more
1041        connectors has changed. Drivers that use the fb helper can just call the
1042        <function>drm_fb_helper_hotplug_event</function> function to handle this
1043        operation.
1044      </para>
1045    </sect2>
1046  </sect1>
1047
1048  <!-- Internals: kms initialization and cleanup -->
1049
1050  <sect1 id="drm-kms-init">
1051    <title>KMS Initialization and Cleanup</title>
1052    <para>
1053      A KMS device is abstracted and exposed as a set of planes, CRTCs, encoders
1054      and connectors. KMS drivers must thus create and initialize all those
1055      objects at load time after initializing mode setting.
1056    </para>
1057    <sect2>
1058      <title>CRTCs (struct <structname>drm_crtc</structname>)</title>
1059      <para>
1060        A CRTC is an abstraction representing a part of the chip that contains a
1061        pointer to a scanout buffer. Therefore, the number of CRTCs available
1062        determines how many independent scanout buffers can be active at any
1063        given time. The CRTC structure contains several fields to support this:
1064        a pointer to some video memory (abstracted as a frame buffer object), a
1065        display mode, and an (x, y) offset into the video memory to support
1066        panning or configurations where one piece of video memory spans multiple
1067        CRTCs.
1068      </para>
1069      <sect3>
1070        <title>CRTC Initialization</title>
1071        <para>
1072          A KMS device must create and register at least one struct
1073          <structname>drm_crtc</structname> instance. The instance is allocated
1074          and zeroed by the driver, possibly as part of a larger structure, and
1075          registered with a call to <function>drm_crtc_init</function> with a
1076          pointer to CRTC functions.
1077        </para>
1078      </sect3>
1079      <sect3>
1080        <title>CRTC Operations</title>
1081        <sect4>
1082          <title>Set Configuration</title>
1083          <synopsis>int (*set_config)(struct drm_mode_set *set);</synopsis>
1084          <para>
1085            Apply a new CRTC configuration to the device. The configuration
1086            specifies a CRTC, a frame buffer to scan out from, a (x,y) position in
1087            the frame buffer, a display mode and an array of connectors to drive
1088            with the CRTC if possible.
1089          </para>
1090          <para>
1091            If the frame buffer specified in the configuration is NULL, the driver
1092            must detach all encoders connected to the CRTC and all connectors
1093            attached to those encoders and disable them.
1094          </para>
1095          <para>
1096            This operation is called with the mode config lock held.
1097          </para>
1098          <note><para>
1099            FIXME: How should set_config interact with DPMS? If the CRTC is
1100            suspended, should it be resumed?
1101          </para></note>
1102        </sect4>
1103        <sect4>
1104          <title>Page Flipping</title>
1105          <synopsis>int (*page_flip)(struct drm_crtc *crtc, struct drm_framebuffer *fb,
1106                   struct drm_pending_vblank_event *event);</synopsis>
1107          <para>
1108            Schedule a page flip to the given frame buffer for the CRTC. This
1109            operation is called with the mode config mutex held.
1110          </para>
1111          <para>
1112            Page flipping is a synchronization mechanism that replaces the frame
1113            buffer being scanned out by the CRTC with a new frame buffer during
1114            vertical blanking, avoiding tearing. When an application requests a page
1115            flip the DRM core verifies that the new frame buffer is large enough to
1116            be scanned out by  the CRTC in the currently configured mode and then
1117            calls the CRTC <methodname>page_flip</methodname> operation with a
1118            pointer to the new frame buffer.
1119          </para>
1120          <para>
1121            The <methodname>page_flip</methodname> operation schedules a page flip.
1122            Once any pending rendering targetting the new frame buffer has
1123            completed, the CRTC will be reprogrammed to display that frame buffer
1124            after the next vertical refresh. The operation must return immediately
1125            without waiting for rendering or page flip to complete and must block
1126            any new rendering to the frame buffer until the page flip completes.
1127          </para>
1128          <para>
1129            If a page flip is already pending, the
1130            <methodname>page_flip</methodname> operation must return
1131            -<errorname>EBUSY</errorname>.
1132          </para>
1133          <para>
1134            To synchronize page flip to vertical blanking the driver will likely
1135            need to enable vertical blanking interrupts. It should call
1136            <function>drm_vblank_get</function> for that purpose, and call
1137            <function>drm_vblank_put</function> after the page flip completes.
1138          </para>
1139          <para>
1140            If the application has requested to be notified when page flip completes
1141            the <methodname>page_flip</methodname> operation will be called with a
1142            non-NULL <parameter>event</parameter> argument pointing to a
1143            <structname>drm_pending_vblank_event</structname> instance. Upon page
1144            flip completion the driver must fill the
1145            <parameter>event</parameter>::<structfield>event</structfield>
1146            <structfield>sequence</structfield>, <structfield>tv_sec</structfield>
1147            and <structfield>tv_usec</structfield> fields with the associated
1148            vertical blanking count and timestamp, add the event to the
1149            <parameter>drm_file</parameter> list of events to be signaled, and wake
1150            up any waiting process. This can be performed with
1151            <programlisting><![CDATA[
1152            struct timeval now;
1153
1154            event->event.sequence = drm_vblank_count_and_time(..., &now);
1155            event->event.tv_sec = now.tv_sec;
1156            event->event.tv_usec = now.tv_usec;
1157
1158            spin_lock_irqsave(&dev->event_lock, flags);
1159            list_add_tail(&event->base.link, &event->base.file_priv->event_list);
1160            wake_up_interruptible(&event->base.file_priv->event_wait);
1161            spin_unlock_irqrestore(&dev->event_lock, flags);
1162            ]]></programlisting>
1163          </para>
1164          <note><para>
1165            FIXME: Could drivers that don't need to wait for rendering to complete
1166            just add the event to <literal>dev-&gt;vblank_event_list</literal> and
1167            let the DRM core handle everything, as for "normal" vertical blanking
1168            events?
1169          </para></note>
1170          <para>
1171            While waiting for the page flip to complete, the
1172            <literal>event-&gt;base.link</literal> list head can be used freely by
1173            the driver to store the pending event in a driver-specific list.
1174          </para>
1175          <para>
1176            If the file handle is closed before the event is signaled, drivers must
1177            take care to destroy the event in their
1178            <methodname>preclose</methodname> operation (and, if needed, call
1179            <function>drm_vblank_put</function>).
1180          </para>
1181        </sect4>
1182        <sect4>
1183          <title>Miscellaneous</title>
1184          <itemizedlist>
1185            <listitem>
1186              <synopsis>void (*gamma_set)(struct drm_crtc *crtc, u16 *r, u16 *g, u16 *b,
1187                        uint32_t start, uint32_t size);</synopsis>
1188              <para>
1189                Apply a gamma table to the device. The operation is optional.
1190              </para>
1191            </listitem>
1192            <listitem>
1193              <synopsis>void (*destroy)(struct drm_crtc *crtc);</synopsis>
1194              <para>
1195                Destroy the CRTC when not needed anymore. See
1196                <xref linkend="drm-kms-init"/>.
1197              </para>
1198            </listitem>
1199          </itemizedlist>
1200        </sect4>
1201      </sect3>
1202    </sect2>
1203    <sect2>
1204      <title>Planes (struct <structname>drm_plane</structname>)</title>
1205      <para>
1206        A plane represents an image source that can be blended with or overlayed
1207        on top of a CRTC during the scanout process. Planes are associated with
1208        a frame buffer to crop a portion of the image memory (source) and
1209        optionally scale it to a destination size. The result is then blended
1210        with or overlayed on top of a CRTC.
1211      </para>
1212      <sect3>
1213        <title>Plane Initialization</title>
1214        <para>
1215          Planes are optional. To create a plane, a KMS drivers allocates and
1216          zeroes an instances of struct <structname>drm_plane</structname>
1217          (possibly as part of a larger structure) and registers it with a call
1218          to <function>drm_plane_init</function>. The function takes a bitmask
1219          of the CRTCs that can be associated with the plane, a pointer to the
1220          plane functions and a list of format supported formats.
1221        </para>
1222      </sect3>
1223      <sect3>
1224        <title>Plane Operations</title>
1225        <itemizedlist>
1226          <listitem>
1227            <synopsis>int (*update_plane)(struct drm_plane *plane, struct drm_crtc *crtc,
1228                        struct drm_framebuffer *fb, int crtc_x, int crtc_y,
1229                        unsigned int crtc_w, unsigned int crtc_h,
1230                        uint32_t src_x, uint32_t src_y,
1231                        uint32_t src_w, uint32_t src_h);</synopsis>
1232            <para>
1233              Enable and configure the plane to use the given CRTC and frame buffer.
1234            </para>
1235            <para>
1236              The source rectangle in frame buffer memory coordinates is given by
1237              the <parameter>src_x</parameter>, <parameter>src_y</parameter>,
1238              <parameter>src_w</parameter> and <parameter>src_h</parameter>
1239              parameters (as 16.16 fixed point values). Devices that don't support
1240              subpixel plane coordinates can ignore the fractional part.
1241            </para>
1242            <para>
1243              The destination rectangle in CRTC coordinates is given by the
1244              <parameter>crtc_x</parameter>, <parameter>crtc_y</parameter>,
1245              <parameter>crtc_w</parameter> and <parameter>crtc_h</parameter>
1246              parameters (as integer values). Devices scale the source rectangle to
1247              the destination rectangle. If scaling is not supported, and the source
1248              rectangle size doesn't match the destination rectangle size, the
1249              driver must return a -<errorname>EINVAL</errorname> error.
1250            </para>
1251          </listitem>
1252          <listitem>
1253            <synopsis>int (*disable_plane)(struct drm_plane *plane);</synopsis>
1254            <para>
1255              Disable the plane. The DRM core calls this method in response to a
1256              DRM_IOCTL_MODE_SETPLANE ioctl call with the frame buffer ID set to 0.
1257              Disabled planes must not be processed by the CRTC.
1258            </para>
1259          </listitem>
1260          <listitem>
1261            <synopsis>void (*destroy)(struct drm_plane *plane);</synopsis>
1262            <para>
1263              Destroy the plane when not needed anymore. See
1264              <xref linkend="drm-kms-init"/>.
1265            </para>
1266          </listitem>
1267        </itemizedlist>
1268      </sect3>
1269    </sect2>
1270    <sect2>
1271      <title>Encoders (struct <structname>drm_encoder</structname>)</title>
1272      <para>
1273        An encoder takes pixel data from a CRTC and converts it to a format
1274        suitable for any attached connectors. On some devices, it may be
1275        possible to have a CRTC send data to more than one encoder. In that
1276        case, both encoders would receive data from the same scanout buffer,
1277        resulting in a "cloned" display configuration across the connectors
1278        attached to each encoder.
1279      </para>
1280      <sect3>
1281        <title>Encoder Initialization</title>
1282        <para>
1283          As for CRTCs, a KMS driver must create, initialize and register at
1284          least one struct <structname>drm_encoder</structname> instance. The
1285          instance is allocated and zeroed by the driver, possibly as part of a
1286          larger structure.
1287        </para>
1288        <para>
1289          Drivers must initialize the struct <structname>drm_encoder</structname>
1290          <structfield>possible_crtcs</structfield> and
1291          <structfield>possible_clones</structfield> fields before registering the
1292          encoder. Both fields are bitmasks of respectively the CRTCs that the
1293          encoder can be connected to, and sibling encoders candidate for cloning.
1294        </para>
1295        <para>
1296          After being initialized, the encoder must be registered with a call to
1297          <function>drm_encoder_init</function>. The function takes a pointer to
1298          the encoder functions and an encoder type. Supported types are
1299          <itemizedlist>
1300            <listitem>
1301              DRM_MODE_ENCODER_DAC for VGA and analog on DVI-I/DVI-A
1302              </listitem>
1303            <listitem>
1304              DRM_MODE_ENCODER_TMDS for DVI, HDMI and (embedded) DisplayPort
1305            </listitem>
1306            <listitem>
1307              DRM_MODE_ENCODER_LVDS for display panels
1308            </listitem>
1309            <listitem>
1310              DRM_MODE_ENCODER_TVDAC for TV output (Composite, S-Video, Component,
1311              SCART)
1312            </listitem>
1313            <listitem>
1314              DRM_MODE_ENCODER_VIRTUAL for virtual machine displays
1315            </listitem>
1316          </itemizedlist>
1317        </para>
1318        <para>
1319          Encoders must be attached to a CRTC to be used. DRM drivers leave
1320          encoders unattached at initialization time. Applications (or the fbdev
1321          compatibility layer when implemented) are responsible for attaching the
1322          encoders they want to use to a CRTC.
1323        </para>
1324      </sect3>
1325      <sect3>
1326        <title>Encoder Operations</title>
1327        <itemizedlist>
1328          <listitem>
1329            <synopsis>void (*destroy)(struct drm_encoder *encoder);</synopsis>
1330            <para>
1331              Called to destroy the encoder when not needed anymore. See
1332              <xref linkend="drm-kms-init"/>.
1333            </para>
1334          </listitem>
1335        </itemizedlist>
1336      </sect3>
1337    </sect2>
1338    <sect2>
1339      <title>Connectors (struct <structname>drm_connector</structname>)</title>
1340      <para>
1341        A connector is the final destination for pixel data on a device, and
1342        usually connects directly to an external display device like a monitor
1343        or laptop panel. A connector can only be attached to one encoder at a
1344        time. The connector is also the structure where information about the
1345        attached display is kept, so it contains fields for display data, EDID
1346        data, DPMS &amp; connection status, and information about modes
1347        supported on the attached displays.
1348      </para>
1349      <sect3>
1350        <title>Connector Initialization</title>
1351        <para>
1352          Finally a KMS driver must create, initialize, register and attach at
1353          least one struct <structname>drm_connector</structname> instance. The
1354          instance is created as other KMS objects and initialized by setting the
1355          following fields.
1356        </para>
1357        <variablelist>
1358          <varlistentry>
1359            <term><structfield>interlace_allowed</structfield></term>
1360            <listitem><para>
1361              Whether the connector can handle interlaced modes.
1362            </para></listitem>
1363          </varlistentry>
1364          <varlistentry>
1365            <term><structfield>doublescan_allowed</structfield></term>
1366            <listitem><para>
1367              Whether the connector can handle doublescan.
1368            </para></listitem>
1369          </varlistentry>
1370          <varlistentry>
1371            <term><structfield>display_info
1372            </structfield></term>
1373            <listitem><para>
1374              Display information is filled from EDID information when a display
1375              is detected. For non hot-pluggable displays such as flat panels in
1376              embedded systems, the driver should initialize the
1377              <structfield>display_info</structfield>.<structfield>width_mm</structfield>
1378              and
1379              <structfield>display_info</structfield>.<structfield>height_mm</structfield>
1380              fields with the physical size of the display.
1381            </para></listitem>
1382          </varlistentry>
1383          <varlistentry>
1384            <term id="drm-kms-connector-polled"><structfield>polled</structfield></term>
1385            <listitem><para>
1386              Connector polling mode, a combination of
1387              <variablelist>
1388                <varlistentry>
1389                  <term>DRM_CONNECTOR_POLL_HPD</term>
1390                  <listitem><para>
1391                    The connector generates hotplug events and doesn't need to be
1392                    periodically polled. The CONNECT and DISCONNECT flags must not
1393                    be set together with the HPD flag.
1394                  </para></listitem>
1395                </varlistentry>
1396                <varlistentry>
1397                  <term>DRM_CONNECTOR_POLL_CONNECT</term>
1398                  <listitem><para>
1399                    Periodically poll the connector for connection.
1400                  </para></listitem>
1401                </varlistentry>
1402                <varlistentry>
1403                  <term>DRM_CONNECTOR_POLL_DISCONNECT</term>
1404                  <listitem><para>
1405                    Periodically poll the connector for disconnection.
1406                  </para></listitem>
1407                </varlistentry>
1408              </variablelist>
1409              Set to 0 for connectors that don't support connection status
1410              discovery.
1411            </para></listitem>
1412          </varlistentry>
1413        </variablelist>
1414        <para>
1415          The connector is then registered with a call to
1416          <function>drm_connector_init</function> with a pointer to the connector
1417          functions and a connector type, and exposed through sysfs with a call to
1418          <function>drm_sysfs_connector_add</function>.
1419        </para>
1420        <para>
1421          Supported connector types are
1422          <itemizedlist>
1423            <listitem>DRM_MODE_CONNECTOR_VGA</listitem>
1424            <listitem>DRM_MODE_CONNECTOR_DVII</listitem>
1425            <listitem>DRM_MODE_CONNECTOR_DVID</listitem>
1426            <listitem>DRM_MODE_CONNECTOR_DVIA</listitem>
1427            <listitem>DRM_MODE_CONNECTOR_Composite</listitem>
1428            <listitem>DRM_MODE_CONNECTOR_SVIDEO</listitem>
1429            <listitem>DRM_MODE_CONNECTOR_LVDS</listitem>
1430            <listitem>DRM_MODE_CONNECTOR_Component</listitem>
1431            <listitem>DRM_MODE_CONNECTOR_9PinDIN</listitem>
1432            <listitem>DRM_MODE_CONNECTOR_DisplayPort</listitem>
1433            <listitem>DRM_MODE_CONNECTOR_HDMIA</listitem>
1434            <listitem>DRM_MODE_CONNECTOR_HDMIB</listitem>
1435            <listitem>DRM_MODE_CONNECTOR_TV</listitem>
1436            <listitem>DRM_MODE_CONNECTOR_eDP</listitem>
1437            <listitem>DRM_MODE_CONNECTOR_VIRTUAL</listitem>
1438          </itemizedlist>
1439        </para>
1440        <para>
1441          Connectors must be attached to an encoder to be used. For devices that
1442          map connectors to encoders 1:1, the connector should be attached at
1443          initialization time with a call to
1444          <function>drm_mode_connector_attach_encoder</function>. The driver must
1445          also set the <structname>drm_connector</structname>
1446          <structfield>encoder</structfield> field to point to the attached
1447          encoder.
1448        </para>
1449        <para>
1450          Finally, drivers must initialize the connectors state change detection
1451          with a call to <function>drm_kms_helper_poll_init</function>. If at
1452          least one connector is pollable but can't generate hotplug interrupts
1453          (indicated by the DRM_CONNECTOR_POLL_CONNECT and
1454          DRM_CONNECTOR_POLL_DISCONNECT connector flags), a delayed work will
1455          automatically be queued to periodically poll for changes. Connectors
1456          that can generate hotplug interrupts must be marked with the
1457          DRM_CONNECTOR_POLL_HPD flag instead, and their interrupt handler must
1458          call <function>drm_helper_hpd_irq_event</function>. The function will
1459          queue a delayed work to check the state of all connectors, but no
1460          periodic polling will be done.
1461        </para>
1462      </sect3>
1463      <sect3>
1464        <title>Connector Operations</title>
1465        <note><para>
1466          Unless otherwise state, all operations are mandatory.
1467        </para></note>
1468        <sect4>
1469          <title>DPMS</title>
1470          <synopsis>void (*dpms)(struct drm_connector *connector, int mode);</synopsis>
1471          <para>
1472            The DPMS operation sets the power state of a connector. The mode
1473            argument is one of
1474            <itemizedlist>
1475              <listitem><para>DRM_MODE_DPMS_ON</para></listitem>
1476              <listitem><para>DRM_MODE_DPMS_STANDBY</para></listitem>
1477              <listitem><para>DRM_MODE_DPMS_SUSPEND</para></listitem>
1478              <listitem><para>DRM_MODE_DPMS_OFF</para></listitem>
1479            </itemizedlist>
1480          </para>
1481          <para>
1482            In all but DPMS_ON mode the encoder to which the connector is attached
1483            should put the display in low-power mode by driving its signals
1484            appropriately. If more than one connector is attached to the encoder
1485            care should be taken not to change the power state of other displays as
1486            a side effect. Low-power mode should be propagated to the encoders and
1487            CRTCs when all related connectors are put in low-power mode.
1488          </para>
1489        </sect4>
1490        <sect4>
1491          <title>Modes</title>
1492          <synopsis>int (*fill_modes)(struct drm_connector *connector, uint32_t max_width,
1493                      uint32_t max_height);</synopsis>
1494          <para>
1495            Fill the mode list with all supported modes for the connector. If the
1496            <parameter>max_width</parameter> and <parameter>max_height</parameter>
1497            arguments are non-zero, the implementation must ignore all modes wider
1498            than <parameter>max_width</parameter> or higher than
1499            <parameter>max_height</parameter>.
1500          </para>
1501          <para>
1502            The connector must also fill in this operation its
1503            <structfield>display_info</structfield>
1504            <structfield>width_mm</structfield> and
1505            <structfield>height_mm</structfield> fields with the connected display
1506            physical size in millimeters. The fields should be set to 0 if the value
1507            isn't known or is not applicable (for instance for projector devices).
1508          </para>
1509        </sect4>
1510        <sect4>
1511          <title>Connection Status</title>
1512          <para>
1513            The connection status is updated through polling or hotplug events when
1514            supported (see <xref linkend="drm-kms-connector-polled"/>). The status
1515            value is reported to userspace through ioctls and must not be used
1516            inside the driver, as it only gets initialized by a call to
1517            <function>drm_mode_getconnector</function> from userspace.
1518          </para>
1519          <synopsis>enum drm_connector_status (*detect)(struct drm_connector *connector,
1520                                        bool force);</synopsis>
1521          <para>
1522            Check to see if anything is attached to the connector. The
1523            <parameter>force</parameter> parameter is set to false whilst polling or
1524            to true when checking the connector due to user request.
1525            <parameter>force</parameter> can be used by the driver to avoid
1526            expensive, destructive operations during automated probing.
1527          </para>
1528          <para>
1529            Return connector_status_connected if something is connected to the
1530            connector, connector_status_disconnected if nothing is connected and
1531            connector_status_unknown if the connection state isn't known.
1532          </para>
1533          <para>
1534            Drivers should only return connector_status_connected if the connection
1535            status has really been probed as connected. Connectors that can't detect
1536            the connection status, or failed connection status probes, should return
1537            connector_status_unknown.
1538          </para>
1539        </sect4>
1540        <sect4>
1541          <title>Miscellaneous</title>
1542          <itemizedlist>
1543            <listitem>
1544              <synopsis>void (*destroy)(struct drm_connector *connector);</synopsis>
1545              <para>
1546                Destroy the connector when not needed anymore. See
1547                <xref linkend="drm-kms-init"/>.
1548              </para>
1549            </listitem>
1550          </itemizedlist>
1551        </sect4>
1552      </sect3>
1553    </sect2>
1554    <sect2>
1555      <title>Cleanup</title>
1556      <para>
1557        The DRM core manages its objects' lifetime. When an object is not needed
1558        anymore the core calls its destroy function, which must clean up and
1559        free every resource allocated for the object. Every
1560        <function>drm_*_init</function> call must be matched with a
1561        corresponding <function>drm_*_cleanup</function> call to cleanup CRTCs
1562        (<function>drm_crtc_cleanup</function>), planes
1563        (<function>drm_plane_cleanup</function>), encoders
1564        (<function>drm_encoder_cleanup</function>) and connectors
1565        (<function>drm_connector_cleanup</function>). Furthermore, connectors
1566        that have been added to sysfs must be removed by a call to
1567        <function>drm_sysfs_connector_remove</function> before calling
1568        <function>drm_connector_cleanup</function>.
1569      </para>
1570      <para>
1571        Connectors state change detection must be cleanup up with a call to
1572        <function>drm_kms_helper_poll_fini</function>.
1573      </para>
1574    </sect2>
1575    <sect2>
1576      <title>Output discovery and initialization example</title>
1577      <programlisting><![CDATA[
1578void intel_crt_init(struct drm_device *dev)
1579{
1580        struct drm_connector *connector;
1581        struct intel_output *intel_output;
1582
1583        intel_output = kzalloc(sizeof(struct intel_output), GFP_KERNEL);
1584        if (!intel_output)
1585                return;
1586
1587        connector = &intel_output->base;
1588        drm_connector_init(dev, &intel_output->base,
1589                           &intel_crt_connector_funcs, DRM_MODE_CONNECTOR_VGA);
1590
1591        drm_encoder_init(dev, &intel_output->enc, &intel_crt_enc_funcs,
1592                         DRM_MODE_ENCODER_DAC);
1593
1594        drm_mode_connector_attach_encoder(&intel_output->base,
1595                                          &intel_output->enc);
1596
1597        /* Set up the DDC bus. */
1598        intel_output->ddc_bus = intel_i2c_create(dev, GPIOA, "CRTDDC_A");
1599        if (!intel_output->ddc_bus) {
1600                dev_printk(KERN_ERR, &dev->pdev->dev, "DDC bus registration "
1601                           "failed.\n");
1602                return;
1603        }
1604
1605        intel_output->type = INTEL_OUTPUT_ANALOG;
1606        connector->interlace_allowed = 0;
1607        connector->doublescan_allowed = 0;
1608
1609        drm_encoder_helper_add(&intel_output->enc, &intel_crt_helper_funcs);
1610        drm_connector_helper_add(connector, &intel_crt_connector_helper_funcs);
1611
1612        drm_sysfs_connector_add(connector);
1613}]]></programlisting>
1614      <para>
1615        In the example above (taken from the i915 driver), a CRTC, connector and
1616        encoder combination is created. A device-specific i2c bus is also
1617        created for fetching EDID data and performing monitor detection. Once
1618        the process is complete, the new connector is registered with sysfs to
1619        make its properties available to applications.
1620      </para>
1621    </sect2>
1622  </sect1>
1623
1624  <!-- Internals: mid-layer helper functions -->
1625
1626  <sect1>
1627    <title>Mid-layer Helper Functions</title>
1628    <para>
1629      The CRTC, encoder and connector functions provided by the drivers
1630      implement the DRM API. They're called by the DRM core and ioctl handlers
1631      to handle device state changes and configuration request. As implementing
1632      those functions often requires logic not specific to drivers, mid-layer
1633      helper functions are available to avoid duplicating boilerplate code.
1634    </para>
1635    <para>
1636      The DRM core contains one mid-layer implementation. The mid-layer provides
1637      implementations of several CRTC, encoder and connector functions (called
1638      from the top of the mid-layer) that pre-process requests and call
1639      lower-level functions provided by the driver (at the bottom of the
1640      mid-layer). For instance, the
1641      <function>drm_crtc_helper_set_config</function> function can be used to
1642      fill the struct <structname>drm_crtc_funcs</structname>
1643      <structfield>set_config</structfield> field. When called, it will split
1644      the <methodname>set_config</methodname> operation in smaller, simpler
1645      operations and call the driver to handle them.
1646    </para>
1647    <para>
1648      To use the mid-layer, drivers call <function>drm_crtc_helper_add</function>,
1649      <function>drm_encoder_helper_add</function> and
1650      <function>drm_connector_helper_add</function> functions to install their
1651      mid-layer bottom operations handlers, and fill the
1652      <structname>drm_crtc_funcs</structname>,
1653      <structname>drm_encoder_funcs</structname> and
1654      <structname>drm_connector_funcs</structname> structures with pointers to
1655      the mid-layer top API functions. Installing the mid-layer bottom operation
1656      handlers is best done right after registering the corresponding KMS object.
1657    </para>
1658    <para>
1659      The mid-layer is not split between CRTC, encoder and connector operations.
1660      To use it, a driver must provide bottom functions for all of the three KMS
1661      entities.
1662    </para>
1663    <sect2>
1664      <title>Helper Functions</title>
1665      <itemizedlist>
1666        <listitem>
1667          <synopsis>int drm_crtc_helper_set_config(struct drm_mode_set *set);</synopsis>
1668          <para>
1669            The <function>drm_crtc_helper_set_config</function> helper function
1670            is a CRTC <methodname>set_config</methodname> implementation. It
1671            first tries to locate the best encoder for each connector by calling
1672            the connector <methodname>best_encoder</methodname> helper
1673            operation.
1674          </para>
1675          <para>
1676            After locating the appropriate encoders, the helper function will
1677            call the <methodname>mode_fixup</methodname> encoder and CRTC helper
1678            operations to adjust the requested mode, or reject it completely in
1679            which case an error will be returned to the application. If the new
1680            configuration after mode adjustment is identical to the current
1681            configuration the helper function will return without performing any
1682            other operation.
1683          </para>
1684          <para>
1685            If the adjusted mode is identical to the current mode but changes to
1686            the frame buffer need to be applied, the
1687            <function>drm_crtc_helper_set_config</function> function will call
1688            the CRTC <methodname>mode_set_base</methodname> helper operation. If
1689            the adjusted mode differs from the current mode, or if the
1690            <methodname>mode_set_base</methodname> helper operation is not
1691            provided, the helper function performs a full mode set sequence by
1692            calling the <methodname>prepare</methodname>,
1693            <methodname>mode_set</methodname> and
1694            <methodname>commit</methodname> CRTC and encoder helper operations,
1695            in that order.
1696          </para>
1697        </listitem>
1698        <listitem>
1699          <synopsis>void drm_helper_connector_dpms(struct drm_connector *connector, int mode);</synopsis>
1700          <para>
1701            The <function>drm_helper_connector_dpms</function> helper function
1702            is a connector <methodname>dpms</methodname> implementation that
1703            tracks power state of connectors. To use the function, drivers must
1704            provide <methodname>dpms</methodname> helper operations for CRTCs
1705            and encoders to apply the DPMS state to the device.
1706          </para>
1707          <para>
1708            The mid-layer doesn't track the power state of CRTCs and encoders.
1709            The <methodname>dpms</methodname> helper operations can thus be
1710            called with a mode identical to the currently active mode.
1711          </para>
1712        </listitem>
1713        <listitem>
1714          <synopsis>int drm_helper_probe_single_connector_modes(struct drm_connector *connector,
1715                                            uint32_t maxX, uint32_t maxY);</synopsis>
1716          <para>
1717            The <function>drm_helper_probe_single_connector_modes</function> helper
1718            function is a connector <methodname>fill_modes</methodname>
1719            implementation that updates the connection status for the connector
1720            and then retrieves a list of modes by calling the connector
1721            <methodname>get_modes</methodname> helper operation.
1722          </para>
1723          <para>
1724            The function filters out modes larger than
1725            <parameter>max_width</parameter> and <parameter>max_height</parameter>
1726            if specified. It then calls the connector
1727            <methodname>mode_valid</methodname> helper operation for  each mode in
1728            the probed list to check whether the mode is valid for the connector.
1729          </para>
1730        </listitem>
1731      </itemizedlist>
1732    </sect2>
1733    <sect2>
1734      <title>CRTC Helper Operations</title>
1735      <itemizedlist>
1736        <listitem id="drm-helper-crtc-mode-fixup">
1737          <synopsis>bool (*mode_fixup)(struct drm_crtc *crtc,
1738                       const struct drm_display_mode *mode,
1739                       struct drm_display_mode *adjusted_mode);</synopsis>
1740          <para>
1741            Let CRTCs adjust the requested mode or reject it completely. This
1742            operation returns true if the mode is accepted (possibly after being
1743            adjusted) or false if it is rejected.
1744          </para>
1745          <para>
1746            The <methodname>mode_fixup</methodname> operation should reject the
1747            mode if it can't reasonably use it. The definition of "reasonable"
1748            is currently fuzzy in this context. One possible behaviour would be
1749            to set the adjusted mode to the panel timings when a fixed-mode
1750            panel is used with hardware capable of scaling. Another behaviour
1751            would be to accept any input mode and adjust it to the closest mode
1752            supported by the hardware (FIXME: This needs to be clarified).
1753          </para>
1754        </listitem>
1755        <listitem>
1756          <synopsis>int (*mode_set_base)(struct drm_crtc *crtc, int x, int y,
1757                     struct drm_framebuffer *old_fb)</synopsis>
1758          <para>
1759            Move the CRTC on the current frame buffer (stored in
1760            <literal>crtc-&gt;fb</literal>) to position (x,y). Any of the frame
1761            buffer, x position or y position may have been modified.
1762          </para>
1763          <para>
1764            This helper operation is optional. If not provided, the
1765            <function>drm_crtc_helper_set_config</function> function will fall
1766            back to the <methodname>mode_set</methodname> helper operation.
1767          </para>
1768          <note><para>
1769            FIXME: Why are x and y passed as arguments, as they can be accessed
1770            through <literal>crtc-&gt;x</literal> and
1771            <literal>crtc-&gt;y</literal>?
1772          </para></note>
1773        </listitem>
1774        <listitem>
1775          <synopsis>void (*prepare)(struct drm_crtc *crtc);</synopsis>
1776          <para>
1777            Prepare the CRTC for mode setting. This operation is called after
1778            validating the requested mode. Drivers use it to perform
1779            device-specific operations required before setting the new mode.
1780          </para>
1781        </listitem>
1782        <listitem>
1783          <synopsis>int (*mode_set)(struct drm_crtc *crtc, struct drm_display_mode *mode,
1784                struct drm_display_mode *adjusted_mode, int x, int y,
1785                struct drm_framebuffer *old_fb);</synopsis>
1786          <para>
1787            Set a new mode, position and frame buffer. Depending on the device
1788            requirements, the mode can be stored internally by the driver and
1789            applied in the <methodname>commit</methodname> operation, or
1790            programmed to the hardware immediately.
1791          </para>
1792          <para>
1793            The <methodname>mode_set</methodname> operation returns 0 on success
1794            or a negative error code if an error occurs.
1795          </para>
1796        </listitem>
1797        <listitem>
1798          <synopsis>void (*commit)(struct drm_crtc *crtc);</synopsis>
1799          <para>
1800            Commit a mode. This operation is called after setting the new mode.
1801            Upon return the device must use the new mode and be fully
1802            operational.
1803          </para>
1804        </listitem>
1805      </itemizedlist>
1806    </sect2>
1807    <sect2>
1808      <title>Encoder Helper Operations</title>
1809      <itemizedlist>
1810        <listitem>
1811          <synopsis>bool (*mode_fixup)(struct drm_encoder *encoder,
1812                       const struct drm_display_mode *mode,
1813                       struct drm_display_mode *adjusted_mode);</synopsis>
1814          <note><para>
1815            FIXME: The mode argument be const, but the i915 driver modifies
1816            mode-&gt;clock in <function>intel_dp_mode_fixup</function>.
1817          </para></note>
1818          <para>
1819            Let encoders adjust the requested mode or reject it completely. This
1820            operation returns true if the mode is accepted (possibly after being
1821            adjusted) or false if it is rejected. See the
1822            <link linkend="drm-helper-crtc-mode-fixup">mode_fixup CRTC helper
1823            operation</link> for an explanation of the allowed adjustments.
1824          </para>
1825        </listitem>
1826        <listitem>
1827          <synopsis>void (*prepare)(struct drm_encoder *encoder);</synopsis>
1828          <para>
1829            Prepare the encoder for mode setting. This operation is called after
1830            validating the requested mode. Drivers use it to perform
1831            device-specific operations required before setting the new mode.
1832          </para>
1833        </listitem>
1834        <listitem>
1835          <synopsis>void (*mode_set)(struct drm_encoder *encoder,
1836                 struct drm_display_mode *mode,
1837                 struct drm_display_mode *adjusted_mode);</synopsis>
1838          <para>
1839            Set a new mode. Depending on the device requirements, the mode can
1840            be stored internally by the driver and applied in the
1841            <methodname>commit</methodname> operation, or programmed to the
1842            hardware immediately.
1843          </para>
1844        </listitem>
1845        <listitem>
1846          <synopsis>void (*commit)(struct drm_encoder *encoder);</synopsis>
1847          <para>
1848            Commit a mode. This operation is called after setting the new mode.
1849            Upon return the device must use the new mode and be fully
1850            operational.
1851          </para>
1852        </listitem>
1853      </itemizedlist>
1854    </sect2>
1855    <sect2>
1856      <title>Connector Helper Operations</title>
1857      <itemizedlist>
1858        <listitem>
1859          <synopsis>struct drm_encoder *(*best_encoder)(struct drm_connector *connector);</synopsis>
1860          <para>
1861            Return a pointer to the best encoder for the connecter. Device that
1862            map connectors to encoders 1:1 simply return the pointer to the
1863            associated encoder. This operation is mandatory.
1864          </para>
1865        </listitem>
1866        <listitem>
1867          <synopsis>int (*get_modes)(struct drm_connector *connector);</synopsis>
1868          <para>
1869            Fill the connector's <structfield>probed_modes</structfield> list
1870            by parsing EDID data with <function>drm_add_edid_modes</function> or
1871            calling <function>drm_mode_probed_add</function> directly for every
1872            supported mode and return the number of modes it has detected. This
1873            operation is mandatory.
1874          </para>
1875          <para>
1876            When adding modes manually the driver creates each mode with a call to
1877            <function>drm_mode_create</function> and must fill the following fields.
1878            <itemizedlist>
1879              <listitem>
1880                <synopsis>__u32 type;</synopsis>
1881                <para>
1882                  Mode type bitmask, a combination of
1883                  <variablelist>
1884                    <varlistentry>
1885                      <term>DRM_MODE_TYPE_BUILTIN</term>
1886                      <listitem><para>not used?</para></listitem>
1887                    </varlistentry>
1888                    <varlistentry>
1889                      <term>DRM_MODE_TYPE_CLOCK_C</term>
1890                      <listitem><para>not used?</para></listitem>
1891                    </varlistentry>
1892                    <varlistentry>
1893                      <term>DRM_MODE_TYPE_CRTC_C</term>
1894                      <listitem><para>not used?</para></listitem>
1895                    </varlistentry>
1896                    <varlistentry>
1897                      <term>
1898        DRM_MODE_TYPE_PREFERRED - The preferred mode for the connector
1899                      </term>
1900                      <listitem>
1901                        <para>not used?</para>
1902                      </listitem>
1903                    </varlistentry>
1904                    <varlistentry>
1905                      <term>DRM_MODE_TYPE_DEFAULT</term>
1906                      <listitem><para>not used?</para></listitem>
1907                    </varlistentry>
1908                    <varlistentry>
1909                      <term>DRM_MODE_TYPE_USERDEF</term>
1910                      <listitem><para>not used?</para></listitem>
1911                    </varlistentry>
1912                    <varlistentry>
1913                      <term>DRM_MODE_TYPE_DRIVER</term>
1914                      <listitem>
1915                        <para>
1916                          The mode has been created by the driver (as opposed to
1917                          to user-created modes).
1918                        </para>
1919                      </listitem>
1920                    </varlistentry>
1921                  </variablelist>
1922                  Drivers must set the DRM_MODE_TYPE_DRIVER bit for all modes they
1923                  create, and set the DRM_MODE_TYPE_PREFERRED bit for the preferred
1924                  mode.
1925                </para>
1926              </listitem>
1927              <listitem>
1928                <synopsis>__u32 clock;</synopsis>
1929                <para>Pixel clock frequency in kHz unit</para>
1930              </listitem>
1931              <listitem>
1932                <synopsis>__u16 hdisplay, hsync_start, hsync_end, htotal;
1933    __u16 vdisplay, vsync_start, vsync_end, vtotal;</synopsis>
1934                <para>Horizontal and vertical timing information</para>
1935                <screen><![CDATA[
1936             Active                 Front           Sync           Back
1937             Region                 Porch                          Porch
1938    <-----------------------><----------------><-------------><-------------->
1939
1940      //////////////////////|
1941     ////////////////////// |
1942    //////////////////////  |..................               ................
1943                                               _______________
1944
1945    <----- [hv]display ----->
1946    <------------- [hv]sync_start ------------>
1947    <--------------------- [hv]sync_end --------------------->
1948    <-------------------------------- [hv]total ----------------------------->
1949]]></screen>
1950              </listitem>
1951              <listitem>
1952                <synopsis>__u16 hskew;
1953    __u16 vscan;</synopsis>
1954                <para>Unknown</para>
1955              </listitem>
1956              <listitem>
1957                <synopsis>__u32 flags;</synopsis>
1958                <para>
1959                  Mode flags, a combination of
1960                  <variablelist>
1961                    <varlistentry>
1962                      <term>DRM_MODE_FLAG_PHSYNC</term>
1963                      <listitem><para>
1964                        Horizontal sync is active high
1965                      </para></listitem>
1966                    </varlistentry>
1967                    <varlistentry>
1968                      <term>DRM_MODE_FLAG_NHSYNC</term>
1969                      <listitem><para>
1970                        Horizontal sync is active low
1971                      </para></listitem>
1972                    </varlistentry>
1973                    <varlistentry>
1974                      <term>DRM_MODE_FLAG_PVSYNC</term>
1975                      <listitem><para>
1976                        Vertical sync is active high
1977                      </para></listitem>
1978                    </varlistentry>
1979                    <varlistentry>
1980                      <term>DRM_MODE_FLAG_NVSYNC</term>
1981                      <listitem><para>
1982                        Vertical sync is active low
1983                      </para></listitem>
1984                    </varlistentry>
1985                    <varlistentry>
1986                      <term>DRM_MODE_FLAG_INTERLACE</term>
1987                      <listitem><para>
1988                        Mode is interlaced
1989                      </para></listitem>
1990                    </varlistentry>
1991                    <varlistentry>
1992                      <term>DRM_MODE_FLAG_DBLSCAN</term>
1993                      <listitem><para>
1994                        Mode uses doublescan
1995                      </para></listitem>
1996                    </varlistentry>
1997                    <varlistentry>
1998                      <term>DRM_MODE_FLAG_CSYNC</term>
1999                      <listitem><para>
2000                        Mode uses composite sync
2001                      </para></listitem>
2002                    </varlistentry>
2003                    <varlistentry>
2004                      <term>DRM_MODE_FLAG_PCSYNC</term>
2005                      <listitem><para>
2006                        Composite sync is active high
2007                      </para></listitem>
2008                    </varlistentry>
2009                    <varlistentry>
2010                      <term>DRM_MODE_FLAG_NCSYNC</term>
2011                      <listitem><para>
2012                        Composite sync is active low
2013                      </para></listitem>
2014                    </varlistentry>
2015                    <varlistentry>
2016                      <term>DRM_MODE_FLAG_HSKEW</term>
2017                      <listitem><para>
2018                        hskew provided (not used?)
2019                      </para></listitem>
2020                    </varlistentry>
2021                    <varlistentry>
2022                      <term>DRM_MODE_FLAG_BCAST</term>
2023                      <listitem><para>
2024                        not used?
2025                      </para></listitem>
2026                    </varlistentry>
2027                    <varlistentry>
2028                      <term>DRM_MODE_FLAG_PIXMUX</term>
2029                      <listitem><para>
2030                        not used?
2031                      </para></listitem>
2032                    </varlistentry>
2033                    <varlistentry>
2034                      <term>DRM_MODE_FLAG_DBLCLK</term>
2035                      <listitem><para>
2036                        not used?
2037                      </para></listitem>
2038                    </varlistentry>
2039                    <varlistentry>
2040                      <term>DRM_MODE_FLAG_CLKDIV2</term>
2041                      <listitem><para>
2042                        ?
2043                      </para></listitem>
2044                    </varlistentry>
2045                  </variablelist>
2046                </para>
2047                <para>
2048                  Note that modes marked with the INTERLACE or DBLSCAN flags will be
2049                  filtered out by
2050                  <function>drm_helper_probe_single_connector_modes</function> if
2051                  the connector's <structfield>interlace_allowed</structfield> or
2052                  <structfield>doublescan_allowed</structfield> field is set to 0.
2053                </para>
2054              </listitem>
2055              <listitem>
2056                <synopsis>char name[DRM_DISPLAY_MODE_LEN];</synopsis>
2057                <para>
2058                  Mode name. The driver must call
2059                  <function>drm_mode_set_name</function> to fill the mode name from
2060                  <structfield>hdisplay</structfield>,
2061                  <structfield>vdisplay</structfield> and interlace flag after
2062                  filling the corresponding fields.
2063                </para>
2064              </listitem>
2065            </itemizedlist>
2066          </para>
2067          <para>
2068            The <structfield>vrefresh</structfield> value is computed by
2069            <function>drm_helper_probe_single_connector_modes</function>.
2070          </para>
2071          <para>
2072            When parsing EDID data, <function>drm_add_edid_modes</function> fill the
2073            connector <structfield>display_info</structfield>
2074            <structfield>width_mm</structfield> and
2075            <structfield>height_mm</structfield> fields. When creating modes
2076            manually the <methodname>get_modes</methodname> helper operation must
2077            set the <structfield>display_info</structfield>
2078            <structfield>width_mm</structfield> and
2079            <structfield>height_mm</structfield> fields if they haven't been set
2080            already (for instance at initilization time when a fixed-size panel is
2081            attached to the connector). The mode <structfield>width_mm</structfield>
2082            and <structfield>height_mm</structfield> fields are only used internally
2083            during EDID parsing and should not be set when creating modes manually.
2084          </para>
2085        </listitem>
2086        <listitem>
2087          <synopsis>int (*mode_valid)(struct drm_connector *connector,
2088                  struct drm_display_mode *mode);</synopsis>
2089          <para>
2090            Verify whether a mode is valid for the connector. Return MODE_OK for
2091            supported modes and one of the enum drm_mode_status values (MODE_*)
2092            for unsupported modes. This operation is mandatory.
2093          </para>
2094          <para>
2095            As the mode rejection reason is currently not used beside for
2096            immediately removing the unsupported mode, an implementation can
2097            return MODE_BAD regardless of the exact reason why the mode is not
2098            valid.
2099          </para>
2100          <note><para>
2101            Note that the <methodname>mode_valid</methodname> helper operation is
2102            only called for modes detected by the device, and
2103            <emphasis>not</emphasis> for modes set by the user through the CRTC
2104            <methodname>set_config</methodname> operation.
2105          </para></note>
2106        </listitem>
2107      </itemizedlist>
2108    </sect2>
2109  </sect1>
2110
2111  <!-- Internals: vertical blanking -->
2112
2113  <sect1 id="drm-vertical-blank">
2114    <title>Vertical Blanking</title>
2115    <para>
2116      Vertical blanking plays a major role in graphics rendering. To achieve
2117      tear-free display, users must synchronize page flips and/or rendering to
2118      vertical blanking. The DRM API offers ioctls to perform page flips
2119      synchronized to vertical blanking and wait for vertical blanking.
2120    </para>
2121    <para>
2122      The DRM core handles most of the vertical blanking management logic, which
2123      involves filtering out spurious interrupts, keeping race-free blanking
2124      counters, coping with counter wrap-around and resets and keeping use
2125      counts. It relies on the driver to generate vertical blanking interrupts
2126      and optionally provide a hardware vertical blanking counter. Drivers must
2127      implement the following operations.
2128    </para>
2129    <itemizedlist>
2130      <listitem>
2131        <synopsis>int (*enable_vblank) (struct drm_device *dev, int crtc);
2132void (*disable_vblank) (struct drm_device *dev, int crtc);</synopsis>
2133        <para>
2134          Enable or disable vertical blanking interrupts for the given CRTC.
2135        </para>
2136      </listitem>
2137      <listitem>
2138        <synopsis>u32 (*get_vblank_counter) (struct drm_device *dev, int crtc);</synopsis>
2139        <para>
2140          Retrieve the value of the vertical blanking counter for the given
2141          CRTC. If the hardware maintains a vertical blanking counter its value
2142          should be returned. Otherwise drivers can use the
2143          <function>drm_vblank_count</function> helper function to handle this
2144          operation.
2145        </para>
2146      </listitem>
2147    </itemizedlist>
2148    <para>
2149      Drivers must initialize the vertical blanking handling core with a call to
2150      <function>drm_vblank_init</function> in their
2151      <methodname>load</methodname> operation. The function will set the struct
2152      <structname>drm_device</structname>
2153      <structfield>vblank_disable_allowed</structfield> field to 0. This will
2154      keep vertical blanking interrupts enabled permanently until the first mode
2155      set operation, where <structfield>vblank_disable_allowed</structfield> is
2156      set to 1. The reason behind this is not clear. Drivers can set the field
2157      to 1 after <function>calling drm_vblank_init</function> to make vertical
2158      blanking interrupts dynamically managed from the beginning.
2159    </para>
2160    <para>
2161      Vertical blanking interrupts can be enabled by the DRM core or by drivers
2162      themselves (for instance to handle page flipping operations). The DRM core
2163      maintains a vertical blanking use count to ensure that the interrupts are
2164      not disabled while a user still needs them. To increment the use count,
2165      drivers call <function>drm_vblank_get</function>. Upon return vertical
2166      blanking interrupts are guaranteed to be enabled.
2167    </para>
2168    <para>
2169      To decrement the use count drivers call
2170      <function>drm_vblank_put</function>. Only when the use count drops to zero
2171      will the DRM core disable the vertical blanking interrupts after a delay
2172      by scheduling a timer. The delay is accessible through the vblankoffdelay
2173      module parameter or the <varname>drm_vblank_offdelay</varname> global
2174      variable and expressed in milliseconds. Its default value is 5000 ms.
2175    </para>
2176    <para>
2177      When a vertical blanking interrupt occurs drivers only need to call the
2178      <function>drm_handle_vblank</function> function to account for the
2179      interrupt.
2180    </para>
2181    <para>
2182      Resources allocated by <function>drm_vblank_init</function> must be freed
2183      with a call to <function>drm_vblank_cleanup</function> in the driver
2184      <methodname>unload</methodname> operation handler.
2185    </para>
2186  </sect1>
2187
2188  <!-- Internals: open/close, file operations and ioctls -->
2189
2190  <sect1>
2191    <title>Open/Close, File Operations and IOCTLs</title>
2192    <sect2>
2193      <title>Open and Close</title>
2194      <synopsis>int (*firstopen) (struct drm_device *);
2195void (*lastclose) (struct drm_device *);
2196int (*open) (struct drm_device *, struct drm_file *);
2197void (*preclose) (struct drm_device *, struct drm_file *);
2198void (*postclose) (struct drm_device *, struct drm_file *);</synopsis>
2199      <abstract>Open and close handlers. None of those methods are mandatory.
2200      </abstract>
2201      <para>
2202        The <methodname>firstopen</methodname> method is called by the DRM core
2203        when an application opens a device that has no other opened file handle.
2204        Similarly the <methodname>lastclose</methodname> method is called when
2205        the last application holding a file handle opened on the device closes
2206        it. Both methods are mostly used for UMS (User Mode Setting) drivers to
2207        acquire and release device resources which should be done in the
2208        <methodname>load</methodname> and <methodname>unload</methodname>
2209        methods for KMS drivers.
2210      </para>
2211      <para>
2212        Note that the <methodname>lastclose</methodname> method is also called
2213        at module unload time or, for hot-pluggable devices, when the device is
2214        unplugged. The <methodname>firstopen</methodname> and
2215        <methodname>lastclose</methodname> calls can thus be unbalanced.
2216      </para>
2217      <para>
2218        The <methodname>open</methodname> method is called every time the device
2219        is opened by an application. Drivers can allocate per-file private data
2220        in this method and store them in the struct
2221        <structname>drm_file</structname> <structfield>driver_priv</structfield>
2222        field. Note that the <methodname>open</methodname> method is called
2223        before <methodname>firstopen</methodname>.
2224      </para>
2225      <para>
2226        The close operation is split into <methodname>preclose</methodname> and
2227        <methodname>postclose</methodname> methods. Drivers must stop and
2228        cleanup all per-file operations in the <methodname>preclose</methodname>
2229        method. For instance pending vertical blanking and page flip events must
2230        be cancelled. No per-file operation is allowed on the file handle after
2231        returning from the <methodname>preclose</methodname> method.
2232      </para>
2233      <para>
2234        Finally the <methodname>postclose</methodname> method is called as the
2235        last step of the close operation, right before calling the
2236        <methodname>lastclose</methodname> method if no other open file handle
2237        exists for the device. Drivers that have allocated per-file private data
2238        in the <methodname>open</methodname> method should free it here.
2239      </para>
2240      <para>
2241        The <methodname>lastclose</methodname> method should restore CRTC and
2242        plane properties to default value, so that a subsequent open of the
2243        device will not inherit state from the previous user.
2244      </para>
2245    </sect2>
2246    <sect2>
2247      <title>File Operations</title>
2248      <synopsis>const struct file_operations *fops</synopsis>
2249      <abstract>File operations for the DRM device node.</abstract>
2250      <para>
2251        Drivers must define the file operations structure that forms the DRM
2252        userspace API entry point, even though most of those operations are
2253        implemented in the DRM core. The <methodname>open</methodname>,
2254        <methodname>release</methodname> and <methodname>ioctl</methodname>
2255        operations are handled by
2256        <programlisting>
2257        .owner = THIS_MODULE,
2258        .open = drm_open,
2259        .release = drm_release,
2260        .unlocked_ioctl = drm_ioctl,
2261  #ifdef CONFIG_COMPAT
2262        .compat_ioctl = drm_compat_ioctl,
2263  #endif
2264        </programlisting>
2265      </para>
2266      <para>
2267        Drivers that implement private ioctls that requires 32/64bit
2268        compatibility support must provide their own
2269        <methodname>compat_ioctl</methodname> handler that processes private
2270        ioctls and calls <function>drm_compat_ioctl</function> for core ioctls.
2271      </para>
2272      <para>
2273        The <methodname>read</methodname> and <methodname>poll</methodname>
2274        operations provide support for reading DRM events and polling them. They
2275        are implemented by
2276        <programlisting>
2277        .poll = drm_poll,
2278        .read = drm_read,
2279        .fasync = drm_fasync,
2280        .llseek = no_llseek,
2281        </programlisting>
2282      </para>
2283      <para>
2284        The memory mapping implementation varies depending on how the driver
2285        manages memory. Pre-GEM drivers will use <function>drm_mmap</function>,
2286        while GEM-aware drivers will use <function>drm_gem_mmap</function>. See
2287        <xref linkend="drm-gem"/>.
2288        <programlisting>
2289        .mmap = drm_gem_mmap,
2290        </programlisting>
2291      </para>
2292      <para>
2293        No other file operation is supported by the DRM API.
2294      </para>
2295    </sect2>
2296    <sect2>
2297      <title>IOCTLs</title>
2298      <synopsis>struct drm_ioctl_desc *ioctls;
2299int num_ioctls;</synopsis>
2300      <abstract>Driver-specific ioctls descriptors table.</abstract>
2301      <para>
2302        Driver-specific ioctls numbers start at DRM_COMMAND_BASE. The ioctls
2303        descriptors table is indexed by the ioctl number offset from the base
2304        value. Drivers can use the DRM_IOCTL_DEF_DRV() macro to initialize the
2305        table entries.
2306      </para>
2307      <para>
2308        <programlisting>DRM_IOCTL_DEF_DRV(ioctl, func, flags)</programlisting>
2309        <para>
2310          <parameter>ioctl</parameter> is the ioctl name. Drivers must define
2311          the DRM_##ioctl and DRM_IOCTL_##ioctl macros to the ioctl number
2312          offset from DRM_COMMAND_BASE and the ioctl number respectively. The
2313          first macro is private to the device while the second must be exposed
2314          to userspace in a public header.
2315        </para>
2316        <para>
2317          <parameter>func</parameter> is a pointer to the ioctl handler function
2318          compatible with the <type>drm_ioctl_t</type> type.
2319          <programlisting>typedef int drm_ioctl_t(struct drm_device *dev, void *data,
2320                struct drm_file *file_priv);</programlisting>
2321        </para>
2322        <para>
2323          <parameter>flags</parameter> is a bitmask combination of the following
2324          values. It restricts how the ioctl is allowed to be called.
2325          <itemizedlist>
2326            <listitem><para>
2327              DRM_AUTH - Only authenticated callers allowed
2328            </para></listitem>
2329            <listitem><para>
2330              DRM_MASTER - The ioctl can only be called on the master file
2331              handle
2332            </para></listitem>
2333            <listitem><para>
2334              DRM_ROOT_ONLY - Only callers with the SYSADMIN capability allowed
2335            </para></listitem>
2336            <listitem><para>
2337              DRM_CONTROL_ALLOW - The ioctl can only be called on a control
2338              device
2339            </para></listitem>
2340            <listitem><para>
2341              DRM_UNLOCKED - The ioctl handler will be called without locking
2342              the DRM global mutex
2343            </para></listitem>
2344          </itemizedlist>
2345        </para>
2346      </para>
2347    </sect2>
2348  </sect1>
2349
2350  <sect1>
2351    <title>Command submission &amp; fencing</title>
2352    <para>
2353      This should cover a few device-specific command submission
2354      implementations.
2355    </para>
2356  </sect1>
2357
2358  <!-- Internals: suspend/resume -->
2359
2360  <sect1>
2361    <title>Suspend/Resume</title>
2362    <para>
2363      The DRM core provides some suspend/resume code, but drivers wanting full
2364      suspend/resume support should provide save() and restore() functions.
2365      These are called at suspend, hibernate, or resume time, and should perform
2366      any state save or restore required by your device across suspend or
2367      hibernate states.
2368    </para>
2369    <synopsis>int (*suspend) (struct drm_device *, pm_message_t state);
2370int (*resume) (struct drm_device *);</synopsis>
2371    <para>
2372      Those are legacy suspend and resume methods. New driver should use the
2373      power management interface provided by their bus type (usually through
2374      the struct <structname>device_driver</structname> dev_pm_ops) and set
2375      these methods to NULL.
2376    </para>
2377  </sect1>
2378
2379  <sect1>
2380    <title>DMA services</title>
2381    <para>
2382      This should cover how DMA mapping etc. is supported by the core.
2383      These functions are deprecated and should not be used.
2384    </para>
2385  </sect1>
2386  </chapter>
2387
2388<!-- TODO
2389
2390- Add a glossary
2391- Document the struct_mutex catch-all lock
2392- Document connector properties
2393
2394- Why is the load method optional?
2395- What are drivers supposed to set the initial display state to, and how?
2396  Connector's DPMS states are not initialized and are thus equal to
2397  DRM_MODE_DPMS_ON. The fbcon compatibility layer calls
2398  drm_helper_disable_unused_functions(), which disables unused encoders and
2399  CRTCs, but doesn't touch the connectors' DPMS state, and
2400  drm_helper_connector_dpms() in reaction to fbdev blanking events. Do drivers
2401  that don't implement (or just don't use) fbcon compatibility need to call
2402  those functions themselves?
2403- KMS drivers must call drm_vblank_pre_modeset() and drm_vblank_post_modeset()
2404  around mode setting. Should this be done in the DRM core?
2405- vblank_disable_allowed is set to 1 in the first drm_vblank_post_modeset()
2406  call and never set back to 0. It seems to be safe to permanently set it to 1
2407  in drm_vblank_init() for KMS driver, and it might be safe for UMS drivers as
2408  well. This should be investigated.
2409- crtc and connector .save and .restore operations are only used internally in
2410  drivers, should they be removed from the core?
2411- encoder mid-layer .save and .restore operations are only used internally in
2412  drivers, should they be removed from the core?
2413- encoder mid-layer .detect operation is only used internally in drivers,
2414  should it be removed from the core?
2415-->
2416
2417  <!-- External interfaces -->
2418
2419  <chapter id="drmExternals">
2420    <title>Userland interfaces</title>
2421    <para>
2422      The DRM core exports several interfaces to applications,
2423      generally intended to be used through corresponding libdrm
2424      wrapper functions.  In addition, drivers export device-specific
2425      interfaces for use by userspace drivers &amp; device-aware
2426      applications through ioctls and sysfs files.
2427    </para>
2428    <para>
2429      External interfaces include: memory mapping, context management,
2430      DMA operations, AGP management, vblank control, fence
2431      management, memory management, and output management.
2432    </para>
2433    <para>
2434      Cover generic ioctls and sysfs layout here.  We only need high-level
2435      info, since man pages should cover the rest.
2436    </para>
2437
2438  <!-- External: vblank handling -->
2439
2440    <sect1>
2441      <title>VBlank event handling</title>
2442      <para>
2443        The DRM core exposes two vertical blank related ioctls:
2444        <variablelist>
2445          <varlistentry>
2446            <term>DRM_IOCTL_WAIT_VBLANK</term>
2447            <listitem>
2448              <para>
2449                This takes a struct drm_wait_vblank structure as its argument,
2450                and it is used to block or request a signal when a specified
2451                vblank event occurs.
2452              </para>
2453            </listitem>
2454          </varlistentry>
2455          <varlistentry>
2456            <term>DRM_IOCTL_MODESET_CTL</term>
2457            <listitem>
2458              <para>
2459                This should be called by application level drivers before and
2460                after mode setting, since on many devices the vertical blank
2461                counter is reset at that time.  Internally, the DRM snapshots
2462                the last vblank count when the ioctl is called with the
2463                _DRM_PRE_MODESET command, so that the counter won't go backwards
2464                (which is dealt with when _DRM_POST_MODESET is used).
2465              </para>
2466            </listitem>
2467          </varlistentry>
2468        </variablelist>
2469<!--!Edrivers/char/drm/drm_irq.c-->
2470      </para>
2471    </sect1>
2472
2473  </chapter>
2474
2475  <!-- API reference -->
2476
2477  <appendix id="drmDriverApi">
2478    <title>DRM Driver API</title>
2479    <para>
2480      Include auto-generated API reference here (need to reference it
2481      from paragraphs above too).
2482    </para>
2483  </appendix>
2484
2485</book>
2486
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.