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" []>
5 <book id="drmDevelopersGuide">
7 <title>Linux DRM Developer's Guide</title>
11 <firstname>Jesse</firstname>
12 <surname>Barnes</surname>
13 <contrib>Initial version</contrib>
15 <orgname>Intel Corporation</orgname>
17 <email>jesse.barnes@intel.com</email>
22 <firstname>Laurent</firstname>
23 <surname>Pinchart</surname>
24 <contrib>Driver internals</contrib>
26 <orgname>Ideas on board SPRL</orgname>
28 <email>laurent.pinchart@ideasonboard.com</email>
33 <firstname>Daniel</firstname>
34 <surname>Vetter</surname>
35 <contrib>Contributions all over the place</contrib>
37 <orgname>Intel Corporation</orgname>
39 <email>daniel.vetter@ffwll.ch</email>
46 <year>2008-2009</year>
47 <year>2013-2014</year>
48 <holder>Intel Corporation</holder>
52 <holder>Laurent Pinchart</holder>
57 The contents of this file may be used under the terms of the GNU
58 General Public License version 2 (the "GPL") as distributed in
59 the kernel source COPYING file.
64 <!-- Put document revisions here, newest first. -->
66 <revnumber>1.0</revnumber>
67 <date>2012-07-13</date>
68 <authorinitials>LP</authorinitials>
69 <revremark>Added extensive documentation about driver internals.
78 <title>DRM Core</title>
81 This first part of the DRM Developer's Guide documents core DRM code,
82 helper libraries for writting drivers and generic userspace interfaces
83 exposed by DRM drivers.
87 <chapter id="drmIntroduction">
88 <title>Introduction</title>
90 The Linux DRM layer contains code intended to support the needs
91 of complex graphics devices, usually containing programmable
92 pipelines well suited to 3D graphics acceleration. Graphics
93 drivers in the kernel may make use of DRM functions to make
94 tasks like memory management, interrupt handling and DMA easier,
95 and provide a uniform interface to applications.
98 A note on versions: this guide covers features found in the DRM
99 tree, including the TTM memory manager, output configuration and
100 mode setting, and the new vblank internals, in addition to all
101 the regular features found in current kernels.
104 [Insert diagram of typical DRM stack here]
110 <chapter id="drmInternals">
111 <title>DRM Internals</title>
113 This chapter documents DRM internals relevant to driver authors
114 and developers working to add support for the latest features to
118 First, we go over some typical driver initialization
119 requirements, like setting up command buffers, creating an
120 initial output configuration, and initializing core services.
121 Subsequent sections cover core internals in more detail,
122 providing implementation notes and examples.
125 The DRM layer provides several services to graphics drivers,
126 many of them driven by the application interfaces it provides
127 through libdrm, the library that wraps most of the DRM ioctls.
128 These include vblank event handling, memory
129 management, output management, framebuffer management, command
130 submission & fencing, suspend/resume support, and DMA
134 <!-- Internals: driver init -->
137 <title>Driver Initialization</title>
139 At the core of every DRM driver is a <structname>drm_driver</structname>
140 structure. Drivers typically statically initialize a drm_driver structure,
141 and then pass it to one of the <function>drm_*_init()</function> functions
142 to register it with the DRM subsystem.
145 The <structname>drm_driver</structname> structure contains static
146 information that describes the driver and features it supports, and
147 pointers to methods that the DRM core will call to implement the DRM API.
148 We will first go through the <structname>drm_driver</structname> static
149 information fields, and will then describe individual operations in
150 details as they get used in later sections.
153 <title>Driver Information</title>
155 <title>Driver Features</title>
157 Drivers inform the DRM core about their requirements and supported
158 features by setting appropriate flags in the
159 <structfield>driver_features</structfield> field. Since those flags
160 influence the DRM core behaviour since registration time, most of them
161 must be set to registering the <structname>drm_driver</structname>
164 <synopsis>u32 driver_features;</synopsis>
166 <title>Driver Feature Flags</title>
168 <term>DRIVER_USE_AGP</term>
170 Driver uses AGP interface, the DRM core will manage AGP resources.
174 <term>DRIVER_REQUIRE_AGP</term>
176 Driver needs AGP interface to function. AGP initialization failure
177 will become a fatal error.
181 <term>DRIVER_PCI_DMA</term>
183 Driver is capable of PCI DMA, mapping of PCI DMA buffers to
184 userspace will be enabled. Deprecated.
188 <term>DRIVER_SG</term>
190 Driver can perform scatter/gather DMA, allocation and mapping of
191 scatter/gather buffers will be enabled. Deprecated.
195 <term>DRIVER_HAVE_DMA</term>
197 Driver supports DMA, the userspace DMA API will be supported.
202 <term>DRIVER_HAVE_IRQ</term><term>DRIVER_IRQ_SHARED</term>
204 DRIVER_HAVE_IRQ indicates whether the driver has an IRQ handler
205 managed by the DRM Core. The core will support simple IRQ handler
206 installation when the flag is set. The installation process is
207 described in <xref linkend="drm-irq-registration"/>.</para>
208 <para>DRIVER_IRQ_SHARED indicates whether the device & handler
209 support shared IRQs (note that this is required of PCI drivers).
213 <term>DRIVER_GEM</term>
215 Driver use the GEM memory manager.
219 <term>DRIVER_MODESET</term>
221 Driver supports mode setting interfaces (KMS).
225 <term>DRIVER_PRIME</term>
227 Driver implements DRM PRIME buffer sharing.
231 <term>DRIVER_RENDER</term>
233 Driver supports dedicated render nodes.
239 <title>Major, Minor and Patchlevel</title>
242 int patchlevel;</synopsis>
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.
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
261 <title>Name, Description and Date</title>
262 <synopsis>char *name;
264 char *date;</synopsis>
266 The driver name is printed to the kernel log at initialization time,
267 used for IRQ registration and passed to userspace through
271 The driver description is a purely informative string passed to
272 userspace through the DRM_IOCTL_VERSION ioctl and otherwise unused by
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.
285 <title>Driver Load</title>
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, performing resource allocation and
290 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"/>).
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 & 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.
310 <synopsis>int (*load) (struct drm_device *, unsigned long flags);</synopsis>
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.
320 <title>Driver Private Data</title>
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.
333 <sect3 id="drm-irq-registration">
334 <title>IRQ Registration</title>
336 The DRM core tries to facilitate IRQ handler registration and
337 unregistration by providing <function>drm_irq_install</function> and
338 <function>drm_irq_uninstall</function> functions. Those functions only
339 support a single interrupt per device, devices that use more than one
340 IRQs need to be handled manually.
343 <title>Managed IRQ Registration</title>
345 Both the <function>drm_irq_install</function> and
346 <function>drm_irq_uninstall</function> functions get the device IRQ by
347 calling <function>drm_dev_to_irq</function>. This inline function will
348 call a bus-specific operation to retrieve the IRQ number. For platform
349 devices, <function>platform_get_irq</function>(..., 0) is used to
350 retrieve the IRQ number.
353 <function>drm_irq_install</function> starts by calling the
354 <methodname>irq_preinstall</methodname> driver operation. The operation
355 is optional and must make sure that the interrupt will not get fired by
356 clearing all pending interrupt flags or disabling the interrupt.
359 The IRQ will then be requested by a call to
360 <function>request_irq</function>. If the DRIVER_IRQ_SHARED driver
361 feature flag is set, a shared (IRQF_SHARED) IRQ handler will be
365 The IRQ handler function must be provided as the mandatory irq_handler
366 driver operation. It will get passed directly to
367 <function>request_irq</function> and thus has the same prototype as all
368 IRQ handlers. It will get called with a pointer to the DRM device as the
372 Finally the function calls the optional
373 <methodname>irq_postinstall</methodname> driver operation. The operation
374 usually enables interrupts (excluding the vblank interrupt, which is
375 enabled separately), but drivers may choose to enable/disable interrupts
379 <function>drm_irq_uninstall</function> is similarly used to uninstall an
380 IRQ handler. It starts by waking up all processes waiting on a vblank
381 interrupt to make sure they don't hang, and then calls the optional
382 <methodname>irq_uninstall</methodname> driver operation. The operation
383 must disable all hardware interrupts. Finally the function frees the IRQ
384 by calling <function>free_irq</function>.
388 <title>Manual IRQ Registration</title>
390 Drivers that require multiple interrupt handlers can't use the managed
391 IRQ registration functions. In that case IRQs must be registered and
392 unregistered manually (usually with the <function>request_irq</function>
393 and <function>free_irq</function> functions, or their devm_* equivalent).
396 When manually registering IRQs, drivers must not set the DRIVER_HAVE_IRQ
397 driver feature flag, and must not provide the
398 <methodname>irq_handler</methodname> driver operation. They must set the
399 <structname>drm_device</structname> <structfield>irq_enabled</structfield>
400 field to 1 upon registration of the IRQs, and clear it to 0 after
401 unregistering the IRQs.
406 <title>Memory Manager Initialization</title>
408 Every DRM driver requires a memory manager which must be initialized at
409 load time. DRM currently contains two memory managers, the Translation
410 Table Manager (TTM) and the Graphics Execution Manager (GEM).
411 This document describes the use of the GEM memory manager only. See
412 <xref linkend="drm-memory-management"/> for details.
416 <title>Miscellaneous Device Configuration</title>
418 Another task that may be necessary for PCI devices during configuration
419 is mapping the video BIOS. On many devices, the VBIOS describes device
420 configuration, LCD panel timings (if any), and contains flags indicating
421 device state. Mapping the BIOS can be done using the pci_map_rom() call,
422 a convenience function that takes care of mapping the actual ROM,
423 whether it has been shadowed into memory (typically at address 0xc0000)
424 or exists on the PCI device in the ROM BAR. Note that after the ROM has
425 been mapped and any necessary information has been extracted, it should
426 be unmapped; on many devices, the ROM address decoder is shared with
427 other BARs, so leaving it mapped could cause undesired behaviour like
428 hangs or memory corruption.
429 <!--!Fdrivers/pci/rom.c pci_map_rom-->
435 <!-- Internals: memory management -->
437 <sect1 id="drm-memory-management">
438 <title>Memory management</title>
440 Modern Linux systems require large amount of graphics memory to store
441 frame buffers, textures, vertices and other graphics-related data. Given
442 the very dynamic nature of many of that data, managing graphics memory
443 efficiently is thus crucial for the graphics stack and plays a central
444 role in the DRM infrastructure.
447 The DRM core includes two memory managers, namely Translation Table Maps
448 (TTM) and Graphics Execution Manager (GEM). TTM was the first DRM memory
449 manager to be developed and tried to be a one-size-fits-them all
450 solution. It provides a single userspace API to accommodate the need of
451 all hardware, supporting both Unified Memory Architecture (UMA) devices
452 and devices with dedicated video RAM (i.e. most discrete video cards).
453 This resulted in a large, complex piece of code that turned out to be
454 hard to use for driver development.
457 GEM started as an Intel-sponsored project in reaction to TTM's
458 complexity. Its design philosophy is completely different: instead of
459 providing a solution to every graphics memory-related problems, GEM
460 identified common code between drivers and created a support library to
461 share it. GEM has simpler initialization and execution requirements than
462 TTM, but has no video RAM management capabitilies and is thus limited to
466 <title>The Translation Table Manager (TTM)</title>
468 TTM design background and information belongs here.
471 <title>TTM initialization</title>
472 <warning><para>This section is outdated.</para></warning>
474 Drivers wishing to support TTM must fill out a drm_bo_driver
475 structure. The structure contains several fields with function
476 pointers for initializing the TTM, allocating and freeing memory,
477 waiting for command completion and fence synchronization, and memory
478 migration. See the radeon_ttm.c file for an example of usage.
481 The ttm_global_reference structure is made up of several fields:
484 struct ttm_global_reference {
485 enum ttm_global_types global_type;
488 int (*init) (struct ttm_global_reference *);
489 void (*release) (struct ttm_global_reference *);
493 There should be one global reference structure for your memory
494 manager as a whole, and there will be others for each object
495 created by the memory manager at runtime. Your global TTM should
496 have a type of TTM_GLOBAL_TTM_MEM. The size field for the global
497 object should be sizeof(struct ttm_mem_global), and the init and
498 release hooks should point at your driver-specific init and
499 release routines, which probably eventually call
500 ttm_mem_global_init and ttm_mem_global_release, respectively.
503 Once your global TTM accounting structure is set up and initialized
504 by calling ttm_global_item_ref() on it,
505 you need to create a buffer object TTM to
506 provide a pool for buffer object allocation by clients and the
507 kernel itself. The type of this object should be TTM_GLOBAL_TTM_BO,
508 and its size should be sizeof(struct ttm_bo_global). Again,
509 driver-specific init and release functions may be provided,
510 likely eventually calling ttm_bo_global_init() and
511 ttm_bo_global_release(), respectively. Also, like the previous
512 object, ttm_global_item_ref() is used to create an initial reference
513 count for the TTM, which will call your initialization function.
518 <title>The Graphics Execution Manager (GEM)</title>
520 The GEM design approach has resulted in a memory manager that doesn't
521 provide full coverage of all (or even all common) use cases in its
522 userspace or kernel API. GEM exposes a set of standard memory-related
523 operations to userspace and a set of helper functions to drivers, and let
524 drivers implement hardware-specific operations with their own private API.
527 The GEM userspace API is described in the
528 <ulink url="http://lwn.net/Articles/283798/"><citetitle>GEM - the Graphics
529 Execution Manager</citetitle></ulink> article on LWN. While slightly
530 outdated, the document provides a good overview of the GEM API principles.
531 Buffer allocation and read and write operations, described as part of the
532 common GEM API, are currently implemented using driver-specific ioctls.
535 GEM is data-agnostic. It manages abstract buffer objects without knowing
536 what individual buffers contain. APIs that require knowledge of buffer
537 contents or purpose, such as buffer allocation or synchronization
538 primitives, are thus outside of the scope of GEM and must be implemented
539 using driver-specific ioctls.
542 On a fundamental level, GEM involves several operations:
544 <listitem>Memory allocation and freeing</listitem>
545 <listitem>Command execution</listitem>
546 <listitem>Aperture management at command execution time</listitem>
548 Buffer object allocation is relatively straightforward and largely
549 provided by Linux's shmem layer, which provides memory to back each
553 Device-specific operations, such as command execution, pinning, buffer
554 read & write, mapping, and domain ownership transfers are left to
555 driver-specific ioctls.
558 <title>GEM Initialization</title>
560 Drivers that use GEM must set the DRIVER_GEM bit in the struct
561 <structname>drm_driver</structname>
562 <structfield>driver_features</structfield> field. The DRM core will
563 then automatically initialize the GEM core before calling the
564 <methodname>load</methodname> operation. Behind the scene, this will
565 create a DRM Memory Manager object which provides an address space
566 pool for object allocation.
569 In a KMS configuration, drivers need to allocate and initialize a
570 command ring buffer following core GEM initialization if required by
571 the hardware. UMA devices usually have what is called a "stolen"
572 memory region, which provides space for the initial framebuffer and
573 large, contiguous memory regions required by the device. This space is
574 typically not managed by GEM, and must be initialized separately into
575 its own DRM MM object.
579 <title>GEM Objects Creation</title>
581 GEM splits creation of GEM objects and allocation of the memory that
582 backs them in two distinct operations.
585 GEM objects are represented by an instance of struct
586 <structname>drm_gem_object</structname>. Drivers usually need to extend
587 GEM objects with private information and thus create a driver-specific
588 GEM object structure type that embeds an instance of struct
589 <structname>drm_gem_object</structname>.
592 To create a GEM object, a driver allocates memory for an instance of its
593 specific GEM object type and initializes the embedded struct
594 <structname>drm_gem_object</structname> with a call to
595 <function>drm_gem_object_init</function>. The function takes a pointer to
596 the DRM device, a pointer to the GEM object and the buffer object size
600 GEM uses shmem to allocate anonymous pageable memory.
601 <function>drm_gem_object_init</function> will create an shmfs file of
602 the requested size and store it into the struct
603 <structname>drm_gem_object</structname> <structfield>filp</structfield>
604 field. The memory is used as either main storage for the object when the
605 graphics hardware uses system memory directly or as a backing store
609 Drivers are responsible for the actual physical pages allocation by
610 calling <function>shmem_read_mapping_page_gfp</function> for each page.
611 Note that they can decide to allocate pages when initializing the GEM
612 object, or to delay allocation until the memory is needed (for instance
613 when a page fault occurs as a result of a userspace memory access or
614 when the driver needs to start a DMA transfer involving the memory).
617 Anonymous pageable memory allocation is not always desired, for instance
618 when the hardware requires physically contiguous system memory as is
619 often the case in embedded devices. Drivers can create GEM objects with
620 no shmfs backing (called private GEM objects) by initializing them with
621 a call to <function>drm_gem_private_object_init</function> instead of
622 <function>drm_gem_object_init</function>. Storage for private GEM
623 objects must be managed by drivers.
626 Drivers that do not need to extend GEM objects with private information
627 can call the <function>drm_gem_object_alloc</function> function to
628 allocate and initialize a struct <structname>drm_gem_object</structname>
629 instance. The GEM core will call the optional driver
630 <methodname>gem_init_object</methodname> operation after initializing
631 the GEM object with <function>drm_gem_object_init</function>.
632 <synopsis>int (*gem_init_object) (struct drm_gem_object *obj);</synopsis>
635 No alloc-and-init function exists for private GEM objects.
639 <title>GEM Objects Lifetime</title>
641 All GEM objects are reference-counted by the GEM core. References can be
642 acquired and release by <function>calling drm_gem_object_reference</function>
643 and <function>drm_gem_object_unreference</function> respectively. The
644 caller must hold the <structname>drm_device</structname>
645 <structfield>struct_mutex</structfield> lock. As a convenience, GEM
646 provides the <function>drm_gem_object_reference_unlocked</function> and
647 <function>drm_gem_object_unreference_unlocked</function> functions that
648 can be called without holding the lock.
651 When the last reference to a GEM object is released the GEM core calls
652 the <structname>drm_driver</structname>
653 <methodname>gem_free_object</methodname> operation. That operation is
654 mandatory for GEM-enabled drivers and must free the GEM object and all
655 associated resources.
658 <synopsis>void (*gem_free_object) (struct drm_gem_object *obj);</synopsis>
659 Drivers are responsible for freeing all GEM object resources, including
660 the resources created by the GEM core. If an mmap offset has been
661 created for the object (in which case
662 <structname>drm_gem_object</structname>::<structfield>map_list</structfield>::<structfield>map</structfield>
663 is not NULL) it must be freed by a call to
664 <function>drm_gem_free_mmap_offset</function>. The shmfs backing store
665 must be released by calling <function>drm_gem_object_release</function>
666 (that function can safely be called if no shmfs backing store has been
671 <title>GEM Objects Naming</title>
673 Communication between userspace and the kernel refers to GEM objects
674 using local handles, global names or, more recently, file descriptors.
675 All of those are 32-bit integer values; the usual Linux kernel limits
676 apply to the file descriptors.
679 GEM handles are local to a DRM file. Applications get a handle to a GEM
680 object through a driver-specific ioctl, and can use that handle to refer
681 to the GEM object in other standard or driver-specific ioctls. Closing a
682 DRM file handle frees all its GEM handles and dereferences the
683 associated GEM objects.
686 To create a handle for a GEM object drivers call
687 <function>drm_gem_handle_create</function>. The function takes a pointer
688 to the DRM file and the GEM object and returns a locally unique handle.
689 When the handle is no longer needed drivers delete it with a call to
690 <function>drm_gem_handle_delete</function>. Finally the GEM object
691 associated with a handle can be retrieved by a call to
692 <function>drm_gem_object_lookup</function>.
695 Handles don't take ownership of GEM objects, they only take a reference
696 to the object that will be dropped when the handle is destroyed. To
697 avoid leaking GEM objects, drivers must make sure they drop the
698 reference(s) they own (such as the initial reference taken at object
699 creation time) as appropriate, without any special consideration for the
700 handle. For example, in the particular case of combined GEM object and
701 handle creation in the implementation of the
702 <methodname>dumb_create</methodname> operation, drivers must drop the
703 initial reference to the GEM object before returning the handle.
706 GEM names are similar in purpose to handles but are not local to DRM
707 files. They can be passed between processes to reference a GEM object
708 globally. Names can't be used directly to refer to objects in the DRM
709 API, applications must convert handles to names and names to handles
710 using the DRM_IOCTL_GEM_FLINK and DRM_IOCTL_GEM_OPEN ioctls
711 respectively. The conversion is handled by the DRM core without any
712 driver-specific support.
715 GEM also supports buffer sharing with dma-buf file descriptors through
716 PRIME. GEM-based drivers must use the provided helpers functions to
717 implement the exporting and importing correctly. See <xref linkend="drm-prime-support" />.
718 Since sharing file descriptors is inherently more secure than the
719 easily guessable and global GEM names it is the preferred buffer
720 sharing mechanism. Sharing buffers through GEM names is only supported
721 for legacy userspace. Furthermore PRIME also allows cross-device
722 buffer sharing since it is based on dma-bufs.
725 <sect3 id="drm-gem-objects-mapping">
726 <title>GEM Objects Mapping</title>
728 Because mapping operations are fairly heavyweight GEM favours
729 read/write-like access to buffers, implemented through driver-specific
730 ioctls, over mapping buffers to userspace. However, when random access
731 to the buffer is needed (to perform software rendering for instance),
732 direct access to the object can be more efficient.
735 The mmap system call can't be used directly to map GEM objects, as they
736 don't have their own file handle. Two alternative methods currently
737 co-exist to map GEM objects to userspace. The first method uses a
738 driver-specific ioctl to perform the mapping operation, calling
739 <function>do_mmap</function> under the hood. This is often considered
740 dubious, seems to be discouraged for new GEM-enabled drivers, and will
741 thus not be described here.
744 The second method uses the mmap system call on the DRM file handle.
745 <synopsis>void *mmap(void *addr, size_t length, int prot, int flags, int fd,
746 off_t offset);</synopsis>
747 DRM identifies the GEM object to be mapped by a fake offset passed
748 through the mmap offset argument. Prior to being mapped, a GEM object
749 must thus be associated with a fake offset. To do so, drivers must call
750 <function>drm_gem_create_mmap_offset</function> on the object. The
751 function allocates a fake offset range from a pool and stores the
752 offset divided by PAGE_SIZE in
753 <literal>obj->map_list.hash.key</literal>. Care must be taken not to
754 call <function>drm_gem_create_mmap_offset</function> if a fake offset
755 has already been allocated for the object. This can be tested by
756 <literal>obj->map_list.map</literal> being non-NULL.
759 Once allocated, the fake offset value
760 (<literal>obj->map_list.hash.key << PAGE_SHIFT</literal>)
761 must be passed to the application in a driver-specific way and can then
762 be used as the mmap offset argument.
765 The GEM core provides a helper method <function>drm_gem_mmap</function>
766 to handle object mapping. The method can be set directly as the mmap
767 file operation handler. It will look up the GEM object based on the
768 offset value and set the VMA operations to the
769 <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
770 field. Note that <function>drm_gem_mmap</function> doesn't map memory to
771 userspace, but relies on the driver-provided fault handler to map pages
775 To use <function>drm_gem_mmap</function>, drivers must fill the struct
776 <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
777 field with a pointer to VM operations.
780 <synopsis>struct vm_operations_struct *gem_vm_ops
782 struct vm_operations_struct {
783 void (*open)(struct vm_area_struct * area);
784 void (*close)(struct vm_area_struct * area);
785 int (*fault)(struct vm_area_struct *vma, struct vm_fault *vmf);
789 The <methodname>open</methodname> and <methodname>close</methodname>
790 operations must update the GEM object reference count. Drivers can use
791 the <function>drm_gem_vm_open</function> and
792 <function>drm_gem_vm_close</function> helper functions directly as open
796 The fault operation handler is responsible for mapping individual pages
797 to userspace when a page fault occurs. Depending on the memory
798 allocation scheme, drivers can allocate pages at fault time, or can
799 decide to allocate memory for the GEM object at the time the object is
803 Drivers that want to map the GEM object upfront instead of handling page
804 faults can implement their own mmap file operation handler.
808 <title>Memory Coherency</title>
810 When mapped to the device or used in a command buffer, backing pages
811 for an object are flushed to memory and marked write combined so as to
812 be coherent with the GPU. Likewise, if the CPU accesses an object
813 after the GPU has finished rendering to the object, then the object
814 must be made coherent with the CPU's view of memory, usually involving
815 GPU cache flushing of various kinds. This core CPU<->GPU
816 coherency management is provided by a device-specific ioctl, which
817 evaluates an object's current domain and performs any necessary
818 flushing or synchronization to put the object into the desired
819 coherency domain (note that the object may be busy, i.e. an active
820 render target; in that case, setting the domain blocks the client and
821 waits for rendering to complete before performing any necessary
822 flushing operations).
826 <title>Command Execution</title>
828 Perhaps the most important GEM function for GPU devices is providing a
829 command execution interface to clients. Client programs construct
830 command buffers containing references to previously allocated memory
831 objects, and then submit them to GEM. At that point, GEM takes care to
832 bind all the objects into the GTT, execute the buffer, and provide
833 necessary synchronization between clients accessing the same buffers.
834 This often involves evicting some objects from the GTT and re-binding
835 others (a fairly expensive operation), and providing relocation
836 support which hides fixed GTT offsets from clients. Clients must take
837 care not to submit command buffers that reference more objects than
838 can fit in the GTT; otherwise, GEM will reject them and no rendering
839 will occur. Similarly, if several objects in the buffer require fence
840 registers to be allocated for correct rendering (e.g. 2D blits on
841 pre-965 chips), care must be taken not to require more fence registers
842 than are available to the client. Such resource management should be
843 abstracted from the client in libdrm.
847 <title>GEM Function Reference</title>
848 !Edrivers/gpu/drm/drm_gem.c
852 <title>VMA Offset Manager</title>
853 !Pdrivers/gpu/drm/drm_vma_manager.c vma offset manager
854 !Edrivers/gpu/drm/drm_vma_manager.c
855 !Iinclude/drm/drm_vma_manager.h
857 <sect2 id="drm-prime-support">
858 <title>PRIME Buffer Sharing</title>
860 PRIME is the cross device buffer sharing framework in drm, originally
861 created for the OPTIMUS range of multi-gpu platforms. To userspace
862 PRIME buffers are dma-buf based file descriptors.
865 <title>Overview and Driver Interface</title>
867 Similar to GEM global names, PRIME file descriptors are
868 also used to share buffer objects across processes. They offer
869 additional security: as file descriptors must be explicitly sent over
870 UNIX domain sockets to be shared between applications, they can't be
871 guessed like the globally unique GEM names.
874 Drivers that support the PRIME
875 API must set the DRIVER_PRIME bit in the struct
876 <structname>drm_driver</structname>
877 <structfield>driver_features</structfield> field, and implement the
878 <methodname>prime_handle_to_fd</methodname> and
879 <methodname>prime_fd_to_handle</methodname> operations.
882 <synopsis>int (*prime_handle_to_fd)(struct drm_device *dev,
883 struct drm_file *file_priv, uint32_t handle,
884 uint32_t flags, int *prime_fd);
885 int (*prime_fd_to_handle)(struct drm_device *dev,
886 struct drm_file *file_priv, int prime_fd,
887 uint32_t *handle);</synopsis>
888 Those two operations convert a handle to a PRIME file descriptor and
889 vice versa. Drivers must use the kernel dma-buf buffer sharing framework
890 to manage the PRIME file descriptors. Similar to the mode setting
891 API PRIME is agnostic to the underlying buffer object manager, as
892 long as handles are 32bit unsinged integers.
895 While non-GEM drivers must implement the operations themselves, GEM
896 drivers must use the <function>drm_gem_prime_handle_to_fd</function>
897 and <function>drm_gem_prime_fd_to_handle</function> helper functions.
898 Those helpers rely on the driver
899 <methodname>gem_prime_export</methodname> and
900 <methodname>gem_prime_import</methodname> operations to create a dma-buf
901 instance from a GEM object (dma-buf exporter role) and to create a GEM
902 object from a dma-buf instance (dma-buf importer role).
905 <synopsis>struct dma_buf * (*gem_prime_export)(struct drm_device *dev,
906 struct drm_gem_object *obj,
908 struct drm_gem_object * (*gem_prime_import)(struct drm_device *dev,
909 struct dma_buf *dma_buf);</synopsis>
910 These two operations are mandatory for GEM drivers that support
915 <title>PRIME Helper Functions</title>
916 !Pdrivers/gpu/drm/drm_prime.c PRIME Helpers
920 <title>PRIME Function References</title>
921 !Edrivers/gpu/drm/drm_prime.c
925 <!-- Internals: mode setting -->
927 <sect1 id="drm-mode-setting">
928 <title>Mode Setting</title>
930 Drivers must initialize the mode setting core by calling
931 <function>drm_mode_config_init</function> on the DRM device. The function
932 initializes the <structname>drm_device</structname>
933 <structfield>mode_config</structfield> field and never fails. Once done,
934 mode configuration must be setup by initializing the following fields.
938 <synopsis>int min_width, min_height;
939 int max_width, max_height;</synopsis>
941 Minimum and maximum width and height of the frame buffers in pixel
946 <synopsis>struct drm_mode_config_funcs *funcs;</synopsis>
947 <para>Mode setting functions.</para>
951 <title>Frame Buffer Creation</title>
952 <synopsis>struct drm_framebuffer *(*fb_create)(struct drm_device *dev,
953 struct drm_file *file_priv,
954 struct drm_mode_fb_cmd2 *mode_cmd);</synopsis>
956 Frame buffers are abstract memory objects that provide a source of
957 pixels to scanout to a CRTC. Applications explicitly request the
958 creation of frame buffers through the DRM_IOCTL_MODE_ADDFB(2) ioctls and
959 receive an opaque handle that can be passed to the KMS CRTC control,
960 plane configuration and page flip functions.
963 Frame buffers rely on the underneath memory manager for low-level memory
964 operations. When creating a frame buffer applications pass a memory
965 handle (or a list of memory handles for multi-planar formats) through
966 the <parameter>drm_mode_fb_cmd2</parameter> argument. For drivers using
967 GEM as their userspace buffer management interface this would be a GEM
968 handle. Drivers are however free to use their own backing storage object
969 handles, e.g. vmwgfx directly exposes special TTM handles to userspace
970 and so expects TTM handles in the create ioctl and not GEM handles.
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.
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. Drivers must fill its <structfield>width</structfield>,
982 <structfield>height</structfield>, <structfield>pitches</structfield>,
983 <structfield>offsets</structfield>, <structfield>depth</structfield>,
984 <structfield>bits_per_pixel</structfield> and
985 <structfield>pixel_format</structfield> fields from the values passed
986 through the <parameter>drm_mode_fb_cmd2</parameter> argument. They
987 should call the <function>drm_helper_mode_fill_fb_struct</function>
988 helper function to do so.
992 The initialization of the new framebuffer instance is finalized with a
993 call to <function>drm_framebuffer_init</function> which takes a pointer
994 to DRM frame buffer operations (struct
995 <structname>drm_framebuffer_funcs</structname>). Note that this function
996 publishes the framebuffer and so from this point on it can be accessed
997 concurrently from other threads. Hence it must be the last step in the
998 driver's framebuffer initialization sequence. Frame buffer operations
1002 <synopsis>int (*create_handle)(struct drm_framebuffer *fb,
1003 struct drm_file *file_priv, unsigned int *handle);</synopsis>
1005 Create a handle to the frame buffer underlying memory object. If
1006 the frame buffer uses a multi-plane format, the handle will
1007 reference the memory object associated with the first plane.
1010 Drivers call <function>drm_gem_handle_create</function> to create
1015 <synopsis>void (*destroy)(struct drm_framebuffer *framebuffer);</synopsis>
1017 Destroy the frame buffer object and frees all associated
1018 resources. Drivers must call
1019 <function>drm_framebuffer_cleanup</function> to free resources
1020 allocated by the DRM core for the frame buffer object, and must
1021 make sure to unreference all memory objects associated with the
1022 frame buffer. Handles created by the
1023 <methodname>create_handle</methodname> operation are released by
1028 <synopsis>int (*dirty)(struct drm_framebuffer *framebuffer,
1029 struct drm_file *file_priv, unsigned flags, unsigned color,
1030 struct drm_clip_rect *clips, unsigned num_clips);</synopsis>
1032 This optional operation notifies the driver that a region of the
1033 frame buffer has changed in response to a DRM_IOCTL_MODE_DIRTYFB
1040 The lifetime of a drm framebuffer is controlled with a reference count,
1041 drivers can grab additional references with
1042 <function>drm_framebuffer_reference</function> </para> and drop them
1043 again with <function>drm_framebuffer_unreference</function>. For
1044 driver-private framebuffers for which the last reference is never
1045 dropped (e.g. for the fbdev framebuffer when the struct
1046 <structname>drm_framebuffer</structname> is embedded into the fbdev
1047 helper struct) drivers can manually clean up a framebuffer at module
1049 <function>drm_framebuffer_unregister_private</function>.
1052 <title>Dumb Buffer Objects</title>
1054 The KMS API doesn't standardize backing storage object creation and
1055 leaves it to driver-specific ioctls. Furthermore actually creating a
1056 buffer object even for GEM-based drivers is done through a
1057 driver-specific ioctl - GEM only has a common userspace interface for
1058 sharing and destroying objects. While not an issue for full-fledged
1059 graphics stacks that include device-specific userspace components (in
1060 libdrm for instance), this limit makes DRM-based early boot graphics
1061 unnecessarily complex.
1064 Dumb objects partly alleviate the problem by providing a standard
1065 API to create dumb buffers suitable for scanout, which can then be used
1066 to create KMS frame buffers.
1069 To support dumb objects drivers must implement the
1070 <methodname>dumb_create</methodname>,
1071 <methodname>dumb_destroy</methodname> and
1072 <methodname>dumb_map_offset</methodname> operations.
1076 <synopsis>int (*dumb_create)(struct drm_file *file_priv, struct drm_device *dev,
1077 struct drm_mode_create_dumb *args);</synopsis>
1079 The <methodname>dumb_create</methodname> operation creates a driver
1080 object (GEM or TTM handle) suitable for scanout based on the
1081 width, height and depth from the struct
1082 <structname>drm_mode_create_dumb</structname> argument. It fills the
1083 argument's <structfield>handle</structfield>,
1084 <structfield>pitch</structfield> and <structfield>size</structfield>
1085 fields with a handle for the newly created object and its line
1086 pitch and size in bytes.
1090 <synopsis>int (*dumb_destroy)(struct drm_file *file_priv, struct drm_device *dev,
1091 uint32_t handle);</synopsis>
1093 The <methodname>dumb_destroy</methodname> operation destroys a dumb
1094 object created by <methodname>dumb_create</methodname>.
1098 <synopsis>int (*dumb_map_offset)(struct drm_file *file_priv, struct drm_device *dev,
1099 uint32_t handle, uint64_t *offset);</synopsis>
1101 The <methodname>dumb_map_offset</methodname> operation associates an
1102 mmap fake offset with the object given by the handle and returns
1103 it. Drivers must use the
1104 <function>drm_gem_create_mmap_offset</function> function to
1105 associate the fake offset as described in
1106 <xref linkend="drm-gem-objects-mapping"/>.
1111 Note that dumb objects may not be used for gpu acceleration, as has been
1112 attempted on some ARM embedded platforms. Such drivers really must have
1113 a hardware-specific ioctl to allocate suitable buffer objects.
1117 <title>Output Polling</title>
1118 <synopsis>void (*output_poll_changed)(struct drm_device *dev);</synopsis>
1120 This operation notifies the driver that the status of one or more
1121 connectors has changed. Drivers that use the fb helper can just call the
1122 <function>drm_fb_helper_hotplug_event</function> function to handle this
1127 <title>Locking</title>
1129 Beside some lookup structures with their own locking (which is hidden
1130 behind the interface functions) most of the modeset state is protected
1131 by the <code>dev-<mode_config.lock</code> mutex and additionally
1132 per-crtc locks to allow cursor updates, pageflips and similar operations
1133 to occur concurrently with background tasks like output detection.
1134 Operations which cross domains like a full modeset always grab all
1135 locks. Drivers there need to protect resources shared between crtcs with
1136 additional locking. They also need to be careful to always grab the
1137 relevant crtc locks if a modset functions touches crtc state, e.g. for
1138 load detection (which does only grab the <code>mode_config.lock</code>
1139 to allow concurrent screen updates on live crtcs).
1144 <!-- Internals: kms initialization and cleanup -->
1146 <sect1 id="drm-kms-init">
1147 <title>KMS Initialization and Cleanup</title>
1149 A KMS device is abstracted and exposed as a set of planes, CRTCs, encoders
1150 and connectors. KMS drivers must thus create and initialize all those
1151 objects at load time after initializing mode setting.
1154 <title>CRTCs (struct <structname>drm_crtc</structname>)</title>
1156 A CRTC is an abstraction representing a part of the chip that contains a
1157 pointer to a scanout buffer. Therefore, the number of CRTCs available
1158 determines how many independent scanout buffers can be active at any
1159 given time. The CRTC structure contains several fields to support this:
1160 a pointer to some video memory (abstracted as a frame buffer object), a
1161 display mode, and an (x, y) offset into the video memory to support
1162 panning or configurations where one piece of video memory spans multiple
1166 <title>CRTC Initialization</title>
1168 A KMS device must create and register at least one struct
1169 <structname>drm_crtc</structname> instance. The instance is allocated
1170 and zeroed by the driver, possibly as part of a larger structure, and
1171 registered with a call to <function>drm_crtc_init</function> with a
1172 pointer to CRTC functions.
1176 <title>CRTC Operations</title>
1178 <title>Set Configuration</title>
1179 <synopsis>int (*set_config)(struct drm_mode_set *set);</synopsis>
1181 Apply a new CRTC configuration to the device. The configuration
1182 specifies a CRTC, a frame buffer to scan out from, a (x,y) position in
1183 the frame buffer, a display mode and an array of connectors to drive
1184 with the CRTC if possible.
1187 If the frame buffer specified in the configuration is NULL, the driver
1188 must detach all encoders connected to the CRTC and all connectors
1189 attached to those encoders and disable them.
1192 This operation is called with the mode config lock held.
1195 Note that the drm core has no notion of restoring the mode setting
1196 state after resume, since all resume handling is in the full
1197 responsibility of the driver. The common mode setting helper library
1198 though provides a helper which can be used for this:
1199 <function>drm_helper_resume_force_mode</function>.
1203 <title>Page Flipping</title>
1204 <synopsis>int (*page_flip)(struct drm_crtc *crtc, struct drm_framebuffer *fb,
1205 struct drm_pending_vblank_event *event);</synopsis>
1207 Schedule a page flip to the given frame buffer for the CRTC. This
1208 operation is called with the mode config mutex held.
1211 Page flipping is a synchronization mechanism that replaces the frame
1212 buffer being scanned out by the CRTC with a new frame buffer during
1213 vertical blanking, avoiding tearing. When an application requests a page
1214 flip the DRM core verifies that the new frame buffer is large enough to
1215 be scanned out by the CRTC in the currently configured mode and then
1216 calls the CRTC <methodname>page_flip</methodname> operation with a
1217 pointer to the new frame buffer.
1220 The <methodname>page_flip</methodname> operation schedules a page flip.
1221 Once any pending rendering targeting the new frame buffer has
1222 completed, the CRTC will be reprogrammed to display that frame buffer
1223 after the next vertical refresh. The operation must return immediately
1224 without waiting for rendering or page flip to complete and must block
1225 any new rendering to the frame buffer until the page flip completes.
1228 If a page flip can be successfully scheduled the driver must set the
1229 <code>drm_crtc-<fb</code> field to the new framebuffer pointed to
1230 by <code>fb</code>. This is important so that the reference counting
1231 on framebuffers stays balanced.
1234 If a page flip is already pending, the
1235 <methodname>page_flip</methodname> operation must return
1236 -<errorname>EBUSY</errorname>.
1239 To synchronize page flip to vertical blanking the driver will likely
1240 need to enable vertical blanking interrupts. It should call
1241 <function>drm_vblank_get</function> for that purpose, and call
1242 <function>drm_vblank_put</function> after the page flip completes.
1245 If the application has requested to be notified when page flip completes
1246 the <methodname>page_flip</methodname> operation will be called with a
1247 non-NULL <parameter>event</parameter> argument pointing to a
1248 <structname>drm_pending_vblank_event</structname> instance. Upon page
1249 flip completion the driver must call <methodname>drm_send_vblank_event</methodname>
1250 to fill in the event and send to wake up any waiting processes.
1251 This can be performed with
1252 <programlisting><![CDATA[
1253 spin_lock_irqsave(&dev->event_lock, flags);
1255 drm_send_vblank_event(dev, pipe, event);
1256 spin_unlock_irqrestore(&dev->event_lock, flags);
1257 ]]></programlisting>
1260 FIXME: Could drivers that don't need to wait for rendering to complete
1261 just add the event to <literal>dev->vblank_event_list</literal> and
1262 let the DRM core handle everything, as for "normal" vertical blanking
1266 While waiting for the page flip to complete, the
1267 <literal>event->base.link</literal> list head can be used freely by
1268 the driver to store the pending event in a driver-specific list.
1271 If the file handle is closed before the event is signaled, drivers must
1272 take care to destroy the event in their
1273 <methodname>preclose</methodname> operation (and, if needed, call
1274 <function>drm_vblank_put</function>).
1278 <title>Miscellaneous</title>
1281 <synopsis>void (*set_property)(struct drm_crtc *crtc,
1282 struct drm_property *property, uint64_t value);</synopsis>
1284 Set the value of the given CRTC property to
1285 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1286 for more information about properties.
1290 <synopsis>void (*gamma_set)(struct drm_crtc *crtc, u16 *r, u16 *g, u16 *b,
1291 uint32_t start, uint32_t size);</synopsis>
1293 Apply a gamma table to the device. The operation is optional.
1297 <synopsis>void (*destroy)(struct drm_crtc *crtc);</synopsis>
1299 Destroy the CRTC when not needed anymore. See
1300 <xref linkend="drm-kms-init"/>.
1308 <title>Planes (struct <structname>drm_plane</structname>)</title>
1310 A plane represents an image source that can be blended with or overlayed
1311 on top of a CRTC during the scanout process. Planes are associated with
1312 a frame buffer to crop a portion of the image memory (source) and
1313 optionally scale it to a destination size. The result is then blended
1314 with or overlayed on top of a CRTC.
1317 <title>Plane Initialization</title>
1319 Planes are optional. To create a plane, a KMS drivers allocates and
1320 zeroes an instances of struct <structname>drm_plane</structname>
1321 (possibly as part of a larger structure) and registers it with a call
1322 to <function>drm_plane_init</function>. The function takes a bitmask
1323 of the CRTCs that can be associated with the plane, a pointer to the
1324 plane functions and a list of format supported formats.
1328 <title>Plane Operations</title>
1331 <synopsis>int (*update_plane)(struct drm_plane *plane, struct drm_crtc *crtc,
1332 struct drm_framebuffer *fb, int crtc_x, int crtc_y,
1333 unsigned int crtc_w, unsigned int crtc_h,
1334 uint32_t src_x, uint32_t src_y,
1335 uint32_t src_w, uint32_t src_h);</synopsis>
1337 Enable and configure the plane to use the given CRTC and frame buffer.
1340 The source rectangle in frame buffer memory coordinates is given by
1341 the <parameter>src_x</parameter>, <parameter>src_y</parameter>,
1342 <parameter>src_w</parameter> and <parameter>src_h</parameter>
1343 parameters (as 16.16 fixed point values). Devices that don't support
1344 subpixel plane coordinates can ignore the fractional part.
1347 The destination rectangle in CRTC coordinates is given by the
1348 <parameter>crtc_x</parameter>, <parameter>crtc_y</parameter>,
1349 <parameter>crtc_w</parameter> and <parameter>crtc_h</parameter>
1350 parameters (as integer values). Devices scale the source rectangle to
1351 the destination rectangle. If scaling is not supported, and the source
1352 rectangle size doesn't match the destination rectangle size, the
1353 driver must return a -<errorname>EINVAL</errorname> error.
1357 <synopsis>int (*disable_plane)(struct drm_plane *plane);</synopsis>
1359 Disable the plane. The DRM core calls this method in response to a
1360 DRM_IOCTL_MODE_SETPLANE ioctl call with the frame buffer ID set to 0.
1361 Disabled planes must not be processed by the CRTC.
1365 <synopsis>void (*destroy)(struct drm_plane *plane);</synopsis>
1367 Destroy the plane when not needed anymore. See
1368 <xref linkend="drm-kms-init"/>.
1375 <title>Encoders (struct <structname>drm_encoder</structname>)</title>
1377 An encoder takes pixel data from a CRTC and converts it to a format
1378 suitable for any attached connectors. On some devices, it may be
1379 possible to have a CRTC send data to more than one encoder. In that
1380 case, both encoders would receive data from the same scanout buffer,
1381 resulting in a "cloned" display configuration across the connectors
1382 attached to each encoder.
1385 <title>Encoder Initialization</title>
1387 As for CRTCs, a KMS driver must create, initialize and register at
1388 least one struct <structname>drm_encoder</structname> instance. The
1389 instance is allocated and zeroed by the driver, possibly as part of a
1393 Drivers must initialize the struct <structname>drm_encoder</structname>
1394 <structfield>possible_crtcs</structfield> and
1395 <structfield>possible_clones</structfield> fields before registering the
1396 encoder. Both fields are bitmasks of respectively the CRTCs that the
1397 encoder can be connected to, and sibling encoders candidate for cloning.
1400 After being initialized, the encoder must be registered with a call to
1401 <function>drm_encoder_init</function>. The function takes a pointer to
1402 the encoder functions and an encoder type. Supported types are
1405 DRM_MODE_ENCODER_DAC for VGA and analog on DVI-I/DVI-A
1408 DRM_MODE_ENCODER_TMDS for DVI, HDMI and (embedded) DisplayPort
1411 DRM_MODE_ENCODER_LVDS for display panels
1414 DRM_MODE_ENCODER_TVDAC for TV output (Composite, S-Video, Component,
1418 DRM_MODE_ENCODER_VIRTUAL for virtual machine displays
1423 Encoders must be attached to a CRTC to be used. DRM drivers leave
1424 encoders unattached at initialization time. Applications (or the fbdev
1425 compatibility layer when implemented) are responsible for attaching the
1426 encoders they want to use to a CRTC.
1430 <title>Encoder Operations</title>
1433 <synopsis>void (*destroy)(struct drm_encoder *encoder);</synopsis>
1435 Called to destroy the encoder when not needed anymore. See
1436 <xref linkend="drm-kms-init"/>.
1440 <synopsis>void (*set_property)(struct drm_plane *plane,
1441 struct drm_property *property, uint64_t value);</synopsis>
1443 Set the value of the given plane property to
1444 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1445 for more information about properties.
1452 <title>Connectors (struct <structname>drm_connector</structname>)</title>
1454 A connector is the final destination for pixel data on a device, and
1455 usually connects directly to an external display device like a monitor
1456 or laptop panel. A connector can only be attached to one encoder at a
1457 time. The connector is also the structure where information about the
1458 attached display is kept, so it contains fields for display data, EDID
1459 data, DPMS & connection status, and information about modes
1460 supported on the attached displays.
1463 <title>Connector Initialization</title>
1465 Finally a KMS driver must create, initialize, register and attach at
1466 least one struct <structname>drm_connector</structname> instance. The
1467 instance is created as other KMS objects and initialized by setting the
1472 <term><structfield>interlace_allowed</structfield></term>
1474 Whether the connector can handle interlaced modes.
1478 <term><structfield>doublescan_allowed</structfield></term>
1480 Whether the connector can handle doublescan.
1484 <term><structfield>display_info
1485 </structfield></term>
1487 Display information is filled from EDID information when a display
1488 is detected. For non hot-pluggable displays such as flat panels in
1489 embedded systems, the driver should initialize the
1490 <structfield>display_info</structfield>.<structfield>width_mm</structfield>
1492 <structfield>display_info</structfield>.<structfield>height_mm</structfield>
1493 fields with the physical size of the display.
1497 <term id="drm-kms-connector-polled"><structfield>polled</structfield></term>
1499 Connector polling mode, a combination of
1502 <term>DRM_CONNECTOR_POLL_HPD</term>
1504 The connector generates hotplug events and doesn't need to be
1505 periodically polled. The CONNECT and DISCONNECT flags must not
1506 be set together with the HPD flag.
1510 <term>DRM_CONNECTOR_POLL_CONNECT</term>
1512 Periodically poll the connector for connection.
1516 <term>DRM_CONNECTOR_POLL_DISCONNECT</term>
1518 Periodically poll the connector for disconnection.
1522 Set to 0 for connectors that don't support connection status
1528 The connector is then registered with a call to
1529 <function>drm_connector_init</function> with a pointer to the connector
1530 functions and a connector type, and exposed through sysfs with a call to
1531 <function>drm_sysfs_connector_add</function>.
1534 Supported connector types are
1536 <listitem>DRM_MODE_CONNECTOR_VGA</listitem>
1537 <listitem>DRM_MODE_CONNECTOR_DVII</listitem>
1538 <listitem>DRM_MODE_CONNECTOR_DVID</listitem>
1539 <listitem>DRM_MODE_CONNECTOR_DVIA</listitem>
1540 <listitem>DRM_MODE_CONNECTOR_Composite</listitem>
1541 <listitem>DRM_MODE_CONNECTOR_SVIDEO</listitem>
1542 <listitem>DRM_MODE_CONNECTOR_LVDS</listitem>
1543 <listitem>DRM_MODE_CONNECTOR_Component</listitem>
1544 <listitem>DRM_MODE_CONNECTOR_9PinDIN</listitem>
1545 <listitem>DRM_MODE_CONNECTOR_DisplayPort</listitem>
1546 <listitem>DRM_MODE_CONNECTOR_HDMIA</listitem>
1547 <listitem>DRM_MODE_CONNECTOR_HDMIB</listitem>
1548 <listitem>DRM_MODE_CONNECTOR_TV</listitem>
1549 <listitem>DRM_MODE_CONNECTOR_eDP</listitem>
1550 <listitem>DRM_MODE_CONNECTOR_VIRTUAL</listitem>
1554 Connectors must be attached to an encoder to be used. For devices that
1555 map connectors to encoders 1:1, the connector should be attached at
1556 initialization time with a call to
1557 <function>drm_mode_connector_attach_encoder</function>. The driver must
1558 also set the <structname>drm_connector</structname>
1559 <structfield>encoder</structfield> field to point to the attached
1563 Finally, drivers must initialize the connectors state change detection
1564 with a call to <function>drm_kms_helper_poll_init</function>. If at
1565 least one connector is pollable but can't generate hotplug interrupts
1566 (indicated by the DRM_CONNECTOR_POLL_CONNECT and
1567 DRM_CONNECTOR_POLL_DISCONNECT connector flags), a delayed work will
1568 automatically be queued to periodically poll for changes. Connectors
1569 that can generate hotplug interrupts must be marked with the
1570 DRM_CONNECTOR_POLL_HPD flag instead, and their interrupt handler must
1571 call <function>drm_helper_hpd_irq_event</function>. The function will
1572 queue a delayed work to check the state of all connectors, but no
1573 periodic polling will be done.
1577 <title>Connector Operations</title>
1579 Unless otherwise state, all operations are mandatory.
1583 <synopsis>void (*dpms)(struct drm_connector *connector, int mode);</synopsis>
1585 The DPMS operation sets the power state of a connector. The mode
1588 <listitem><para>DRM_MODE_DPMS_ON</para></listitem>
1589 <listitem><para>DRM_MODE_DPMS_STANDBY</para></listitem>
1590 <listitem><para>DRM_MODE_DPMS_SUSPEND</para></listitem>
1591 <listitem><para>DRM_MODE_DPMS_OFF</para></listitem>
1595 In all but DPMS_ON mode the encoder to which the connector is attached
1596 should put the display in low-power mode by driving its signals
1597 appropriately. If more than one connector is attached to the encoder
1598 care should be taken not to change the power state of other displays as
1599 a side effect. Low-power mode should be propagated to the encoders and
1600 CRTCs when all related connectors are put in low-power mode.
1604 <title>Modes</title>
1605 <synopsis>int (*fill_modes)(struct drm_connector *connector, uint32_t max_width,
1606 uint32_t max_height);</synopsis>
1608 Fill the mode list with all supported modes for the connector. If the
1609 <parameter>max_width</parameter> and <parameter>max_height</parameter>
1610 arguments are non-zero, the implementation must ignore all modes wider
1611 than <parameter>max_width</parameter> or higher than
1612 <parameter>max_height</parameter>.
1615 The connector must also fill in this operation its
1616 <structfield>display_info</structfield>
1617 <structfield>width_mm</structfield> and
1618 <structfield>height_mm</structfield> fields with the connected display
1619 physical size in millimeters. The fields should be set to 0 if the value
1620 isn't known or is not applicable (for instance for projector devices).
1624 <title>Connection Status</title>
1626 The connection status is updated through polling or hotplug events when
1627 supported (see <xref linkend="drm-kms-connector-polled"/>). The status
1628 value is reported to userspace through ioctls and must not be used
1629 inside the driver, as it only gets initialized by a call to
1630 <function>drm_mode_getconnector</function> from userspace.
1632 <synopsis>enum drm_connector_status (*detect)(struct drm_connector *connector,
1633 bool force);</synopsis>
1635 Check to see if anything is attached to the connector. The
1636 <parameter>force</parameter> parameter is set to false whilst polling or
1637 to true when checking the connector due to user request.
1638 <parameter>force</parameter> can be used by the driver to avoid
1639 expensive, destructive operations during automated probing.
1642 Return connector_status_connected if something is connected to the
1643 connector, connector_status_disconnected if nothing is connected and
1644 connector_status_unknown if the connection state isn't known.
1647 Drivers should only return connector_status_connected if the connection
1648 status has really been probed as connected. Connectors that can't detect
1649 the connection status, or failed connection status probes, should return
1650 connector_status_unknown.
1654 <title>Miscellaneous</title>
1657 <synopsis>void (*set_property)(struct drm_connector *connector,
1658 struct drm_property *property, uint64_t value);</synopsis>
1660 Set the value of the given connector property to
1661 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1662 for more information about properties.
1666 <synopsis>void (*destroy)(struct drm_connector *connector);</synopsis>
1668 Destroy the connector when not needed anymore. See
1669 <xref linkend="drm-kms-init"/>.
1677 <title>Cleanup</title>
1679 The DRM core manages its objects' lifetime. When an object is not needed
1680 anymore the core calls its destroy function, which must clean up and
1681 free every resource allocated for the object. Every
1682 <function>drm_*_init</function> call must be matched with a
1683 corresponding <function>drm_*_cleanup</function> call to cleanup CRTCs
1684 (<function>drm_crtc_cleanup</function>), planes
1685 (<function>drm_plane_cleanup</function>), encoders
1686 (<function>drm_encoder_cleanup</function>) and connectors
1687 (<function>drm_connector_cleanup</function>). Furthermore, connectors
1688 that have been added to sysfs must be removed by a call to
1689 <function>drm_sysfs_connector_remove</function> before calling
1690 <function>drm_connector_cleanup</function>.
1693 Connectors state change detection must be cleanup up with a call to
1694 <function>drm_kms_helper_poll_fini</function>.
1698 <title>Output discovery and initialization example</title>
1699 <programlisting><![CDATA[
1700 void intel_crt_init(struct drm_device *dev)
1702 struct drm_connector *connector;
1703 struct intel_output *intel_output;
1705 intel_output = kzalloc(sizeof(struct intel_output), GFP_KERNEL);
1709 connector = &intel_output->base;
1710 drm_connector_init(dev, &intel_output->base,
1711 &intel_crt_connector_funcs, DRM_MODE_CONNECTOR_VGA);
1713 drm_encoder_init(dev, &intel_output->enc, &intel_crt_enc_funcs,
1714 DRM_MODE_ENCODER_DAC);
1716 drm_mode_connector_attach_encoder(&intel_output->base,
1717 &intel_output->enc);
1719 /* Set up the DDC bus. */
1720 intel_output->ddc_bus = intel_i2c_create(dev, GPIOA, "CRTDDC_A");
1721 if (!intel_output->ddc_bus) {
1722 dev_printk(KERN_ERR, &dev->pdev->dev, "DDC bus registration "
1727 intel_output->type = INTEL_OUTPUT_ANALOG;
1728 connector->interlace_allowed = 0;
1729 connector->doublescan_allowed = 0;
1731 drm_encoder_helper_add(&intel_output->enc, &intel_crt_helper_funcs);
1732 drm_connector_helper_add(connector, &intel_crt_connector_helper_funcs);
1734 drm_sysfs_connector_add(connector);
1735 }]]></programlisting>
1737 In the example above (taken from the i915 driver), a CRTC, connector and
1738 encoder combination is created. A device-specific i2c bus is also
1739 created for fetching EDID data and performing monitor detection. Once
1740 the process is complete, the new connector is registered with sysfs to
1741 make its properties available to applications.
1745 <title>KMS API Functions</title>
1746 !Edrivers/gpu/drm/drm_crtc.c
1750 <!-- Internals: kms helper functions -->
1753 <title>Mode Setting Helper Functions</title>
1755 The CRTC, encoder and connector functions provided by the drivers
1756 implement the DRM API. They're called by the DRM core and ioctl handlers
1757 to handle device state changes and configuration request. As implementing
1758 those functions often requires logic not specific to drivers, mid-layer
1759 helper functions are available to avoid duplicating boilerplate code.
1762 The DRM core contains one mid-layer implementation. The mid-layer provides
1763 implementations of several CRTC, encoder and connector functions (called
1764 from the top of the mid-layer) that pre-process requests and call
1765 lower-level functions provided by the driver (at the bottom of the
1766 mid-layer). For instance, the
1767 <function>drm_crtc_helper_set_config</function> function can be used to
1768 fill the struct <structname>drm_crtc_funcs</structname>
1769 <structfield>set_config</structfield> field. When called, it will split
1770 the <methodname>set_config</methodname> operation in smaller, simpler
1771 operations and call the driver to handle them.
1774 To use the mid-layer, drivers call <function>drm_crtc_helper_add</function>,
1775 <function>drm_encoder_helper_add</function> and
1776 <function>drm_connector_helper_add</function> functions to install their
1777 mid-layer bottom operations handlers, and fill the
1778 <structname>drm_crtc_funcs</structname>,
1779 <structname>drm_encoder_funcs</structname> and
1780 <structname>drm_connector_funcs</structname> structures with pointers to
1781 the mid-layer top API functions. Installing the mid-layer bottom operation
1782 handlers is best done right after registering the corresponding KMS object.
1785 The mid-layer is not split between CRTC, encoder and connector operations.
1786 To use it, a driver must provide bottom functions for all of the three KMS
1790 <title>Helper Functions</title>
1793 <synopsis>int drm_crtc_helper_set_config(struct drm_mode_set *set);</synopsis>
1795 The <function>drm_crtc_helper_set_config</function> helper function
1796 is a CRTC <methodname>set_config</methodname> implementation. It
1797 first tries to locate the best encoder for each connector by calling
1798 the connector <methodname>best_encoder</methodname> helper
1802 After locating the appropriate encoders, the helper function will
1803 call the <methodname>mode_fixup</methodname> encoder and CRTC helper
1804 operations to adjust the requested mode, or reject it completely in
1805 which case an error will be returned to the application. If the new
1806 configuration after mode adjustment is identical to the current
1807 configuration the helper function will return without performing any
1811 If the adjusted mode is identical to the current mode but changes to
1812 the frame buffer need to be applied, the
1813 <function>drm_crtc_helper_set_config</function> function will call
1814 the CRTC <methodname>mode_set_base</methodname> helper operation. If
1815 the adjusted mode differs from the current mode, or if the
1816 <methodname>mode_set_base</methodname> helper operation is not
1817 provided, the helper function performs a full mode set sequence by
1818 calling the <methodname>prepare</methodname>,
1819 <methodname>mode_set</methodname> and
1820 <methodname>commit</methodname> CRTC and encoder helper operations,
1825 <synopsis>void drm_helper_connector_dpms(struct drm_connector *connector, int mode);</synopsis>
1827 The <function>drm_helper_connector_dpms</function> helper function
1828 is a connector <methodname>dpms</methodname> implementation that
1829 tracks power state of connectors. To use the function, drivers must
1830 provide <methodname>dpms</methodname> helper operations for CRTCs
1831 and encoders to apply the DPMS state to the device.
1834 The mid-layer doesn't track the power state of CRTCs and encoders.
1835 The <methodname>dpms</methodname> helper operations can thus be
1836 called with a mode identical to the currently active mode.
1840 <synopsis>int drm_helper_probe_single_connector_modes(struct drm_connector *connector,
1841 uint32_t maxX, uint32_t maxY);</synopsis>
1843 The <function>drm_helper_probe_single_connector_modes</function> helper
1844 function is a connector <methodname>fill_modes</methodname>
1845 implementation that updates the connection status for the connector
1846 and then retrieves a list of modes by calling the connector
1847 <methodname>get_modes</methodname> helper operation.
1850 The function filters out modes larger than
1851 <parameter>max_width</parameter> and <parameter>max_height</parameter>
1852 if specified. It then calls the connector
1853 <methodname>mode_valid</methodname> helper operation for each mode in
1854 the probed list to check whether the mode is valid for the connector.
1860 <title>CRTC Helper Operations</title>
1862 <listitem id="drm-helper-crtc-mode-fixup">
1863 <synopsis>bool (*mode_fixup)(struct drm_crtc *crtc,
1864 const struct drm_display_mode *mode,
1865 struct drm_display_mode *adjusted_mode);</synopsis>
1867 Let CRTCs adjust the requested mode or reject it completely. This
1868 operation returns true if the mode is accepted (possibly after being
1869 adjusted) or false if it is rejected.
1872 The <methodname>mode_fixup</methodname> operation should reject the
1873 mode if it can't reasonably use it. The definition of "reasonable"
1874 is currently fuzzy in this context. One possible behaviour would be
1875 to set the adjusted mode to the panel timings when a fixed-mode
1876 panel is used with hardware capable of scaling. Another behaviour
1877 would be to accept any input mode and adjust it to the closest mode
1878 supported by the hardware (FIXME: This needs to be clarified).
1882 <synopsis>int (*mode_set_base)(struct drm_crtc *crtc, int x, int y,
1883 struct drm_framebuffer *old_fb)</synopsis>
1885 Move the CRTC on the current frame buffer (stored in
1886 <literal>crtc->fb</literal>) to position (x,y). Any of the frame
1887 buffer, x position or y position may have been modified.
1890 This helper operation is optional. If not provided, the
1891 <function>drm_crtc_helper_set_config</function> function will fall
1892 back to the <methodname>mode_set</methodname> helper operation.
1895 FIXME: Why are x and y passed as arguments, as they can be accessed
1896 through <literal>crtc->x</literal> and
1897 <literal>crtc->y</literal>?
1901 <synopsis>void (*prepare)(struct drm_crtc *crtc);</synopsis>
1903 Prepare the CRTC for mode setting. This operation is called after
1904 validating the requested mode. Drivers use it to perform
1905 device-specific operations required before setting the new mode.
1909 <synopsis>int (*mode_set)(struct drm_crtc *crtc, struct drm_display_mode *mode,
1910 struct drm_display_mode *adjusted_mode, int x, int y,
1911 struct drm_framebuffer *old_fb);</synopsis>
1913 Set a new mode, position and frame buffer. Depending on the device
1914 requirements, the mode can be stored internally by the driver and
1915 applied in the <methodname>commit</methodname> operation, or
1916 programmed to the hardware immediately.
1919 The <methodname>mode_set</methodname> operation returns 0 on success
1920 or a negative error code if an error occurs.
1924 <synopsis>void (*commit)(struct drm_crtc *crtc);</synopsis>
1926 Commit a mode. This operation is called after setting the new mode.
1927 Upon return the device must use the new mode and be fully
1934 <title>Encoder Helper Operations</title>
1937 <synopsis>bool (*mode_fixup)(struct drm_encoder *encoder,
1938 const struct drm_display_mode *mode,
1939 struct drm_display_mode *adjusted_mode);</synopsis>
1941 Let encoders adjust the requested mode or reject it completely. This
1942 operation returns true if the mode is accepted (possibly after being
1943 adjusted) or false if it is rejected. See the
1944 <link linkend="drm-helper-crtc-mode-fixup">mode_fixup CRTC helper
1945 operation</link> for an explanation of the allowed adjustments.
1949 <synopsis>void (*prepare)(struct drm_encoder *encoder);</synopsis>
1951 Prepare the encoder for mode setting. This operation is called after
1952 validating the requested mode. Drivers use it to perform
1953 device-specific operations required before setting the new mode.
1957 <synopsis>void (*mode_set)(struct drm_encoder *encoder,
1958 struct drm_display_mode *mode,
1959 struct drm_display_mode *adjusted_mode);</synopsis>
1961 Set a new mode. Depending on the device requirements, the mode can
1962 be stored internally by the driver and applied in the
1963 <methodname>commit</methodname> operation, or programmed to the
1964 hardware immediately.
1968 <synopsis>void (*commit)(struct drm_encoder *encoder);</synopsis>
1970 Commit a mode. This operation is called after setting the new mode.
1971 Upon return the device must use the new mode and be fully
1978 <title>Connector Helper Operations</title>
1981 <synopsis>struct drm_encoder *(*best_encoder)(struct drm_connector *connector);</synopsis>
1983 Return a pointer to the best encoder for the connecter. Device that
1984 map connectors to encoders 1:1 simply return the pointer to the
1985 associated encoder. This operation is mandatory.
1989 <synopsis>int (*get_modes)(struct drm_connector *connector);</synopsis>
1991 Fill the connector's <structfield>probed_modes</structfield> list
1992 by parsing EDID data with <function>drm_add_edid_modes</function> or
1993 calling <function>drm_mode_probed_add</function> directly for every
1994 supported mode and return the number of modes it has detected. This
1995 operation is mandatory.
1998 When adding modes manually the driver creates each mode with a call to
1999 <function>drm_mode_create</function> and must fill the following fields.
2002 <synopsis>__u32 type;</synopsis>
2004 Mode type bitmask, a combination of
2007 <term>DRM_MODE_TYPE_BUILTIN</term>
2008 <listitem><para>not used?</para></listitem>
2011 <term>DRM_MODE_TYPE_CLOCK_C</term>
2012 <listitem><para>not used?</para></listitem>
2015 <term>DRM_MODE_TYPE_CRTC_C</term>
2016 <listitem><para>not used?</para></listitem>
2020 DRM_MODE_TYPE_PREFERRED - The preferred mode for the connector
2023 <para>not used?</para>
2027 <term>DRM_MODE_TYPE_DEFAULT</term>
2028 <listitem><para>not used?</para></listitem>
2031 <term>DRM_MODE_TYPE_USERDEF</term>
2032 <listitem><para>not used?</para></listitem>
2035 <term>DRM_MODE_TYPE_DRIVER</term>
2038 The mode has been created by the driver (as opposed to
2039 to user-created modes).
2044 Drivers must set the DRM_MODE_TYPE_DRIVER bit for all modes they
2045 create, and set the DRM_MODE_TYPE_PREFERRED bit for the preferred
2050 <synopsis>__u32 clock;</synopsis>
2051 <para>Pixel clock frequency in kHz unit</para>
2054 <synopsis>__u16 hdisplay, hsync_start, hsync_end, htotal;
2055 __u16 vdisplay, vsync_start, vsync_end, vtotal;</synopsis>
2056 <para>Horizontal and vertical timing information</para>
2058 Active Front Sync Back
2060 <-----------------------><----------------><-------------><-------------->
2062 //////////////////////|
2063 ////////////////////// |
2064 ////////////////////// |.................. ................
2067 <----- [hv]display ----->
2068 <------------- [hv]sync_start ------------>
2069 <--------------------- [hv]sync_end --------------------->
2070 <-------------------------------- [hv]total ----------------------------->
2074 <synopsis>__u16 hskew;
2075 __u16 vscan;</synopsis>
2076 <para>Unknown</para>
2079 <synopsis>__u32 flags;</synopsis>
2081 Mode flags, a combination of
2084 <term>DRM_MODE_FLAG_PHSYNC</term>
2086 Horizontal sync is active high
2090 <term>DRM_MODE_FLAG_NHSYNC</term>
2092 Horizontal sync is active low
2096 <term>DRM_MODE_FLAG_PVSYNC</term>
2098 Vertical sync is active high
2102 <term>DRM_MODE_FLAG_NVSYNC</term>
2104 Vertical sync is active low
2108 <term>DRM_MODE_FLAG_INTERLACE</term>
2114 <term>DRM_MODE_FLAG_DBLSCAN</term>
2116 Mode uses doublescan
2120 <term>DRM_MODE_FLAG_CSYNC</term>
2122 Mode uses composite sync
2126 <term>DRM_MODE_FLAG_PCSYNC</term>
2128 Composite sync is active high
2132 <term>DRM_MODE_FLAG_NCSYNC</term>
2134 Composite sync is active low
2138 <term>DRM_MODE_FLAG_HSKEW</term>
2140 hskew provided (not used?)
2144 <term>DRM_MODE_FLAG_BCAST</term>
2150 <term>DRM_MODE_FLAG_PIXMUX</term>
2156 <term>DRM_MODE_FLAG_DBLCLK</term>
2162 <term>DRM_MODE_FLAG_CLKDIV2</term>
2170 Note that modes marked with the INTERLACE or DBLSCAN flags will be
2172 <function>drm_helper_probe_single_connector_modes</function> if
2173 the connector's <structfield>interlace_allowed</structfield> or
2174 <structfield>doublescan_allowed</structfield> field is set to 0.
2178 <synopsis>char name[DRM_DISPLAY_MODE_LEN];</synopsis>
2180 Mode name. The driver must call
2181 <function>drm_mode_set_name</function> to fill the mode name from
2182 <structfield>hdisplay</structfield>,
2183 <structfield>vdisplay</structfield> and interlace flag after
2184 filling the corresponding fields.
2190 The <structfield>vrefresh</structfield> value is computed by
2191 <function>drm_helper_probe_single_connector_modes</function>.
2194 When parsing EDID data, <function>drm_add_edid_modes</function> fill the
2195 connector <structfield>display_info</structfield>
2196 <structfield>width_mm</structfield> and
2197 <structfield>height_mm</structfield> fields. When creating modes
2198 manually the <methodname>get_modes</methodname> helper operation must
2199 set the <structfield>display_info</structfield>
2200 <structfield>width_mm</structfield> and
2201 <structfield>height_mm</structfield> fields if they haven't been set
2202 already (for instance at initialization time when a fixed-size panel is
2203 attached to the connector). The mode <structfield>width_mm</structfield>
2204 and <structfield>height_mm</structfield> fields are only used internally
2205 during EDID parsing and should not be set when creating modes manually.
2209 <synopsis>int (*mode_valid)(struct drm_connector *connector,
2210 struct drm_display_mode *mode);</synopsis>
2212 Verify whether a mode is valid for the connector. Return MODE_OK for
2213 supported modes and one of the enum drm_mode_status values (MODE_*)
2214 for unsupported modes. This operation is mandatory.
2217 As the mode rejection reason is currently not used beside for
2218 immediately removing the unsupported mode, an implementation can
2219 return MODE_BAD regardless of the exact reason why the mode is not
2223 Note that the <methodname>mode_valid</methodname> helper operation is
2224 only called for modes detected by the device, and
2225 <emphasis>not</emphasis> for modes set by the user through the CRTC
2226 <methodname>set_config</methodname> operation.
2232 <title>Modeset Helper Functions Reference</title>
2233 !Edrivers/gpu/drm/drm_crtc_helper.c
2236 <title>fbdev Helper Functions Reference</title>
2237 !Pdrivers/gpu/drm/drm_fb_helper.c fbdev helpers
2238 !Edrivers/gpu/drm/drm_fb_helper.c
2239 !Iinclude/drm/drm_fb_helper.h
2242 <title>Display Port Helper Functions Reference</title>
2243 !Pdrivers/gpu/drm/drm_dp_helper.c dp helpers
2244 !Iinclude/drm/drm_dp_helper.h
2245 !Edrivers/gpu/drm/drm_dp_helper.c
2248 <title>EDID Helper Functions Reference</title>
2249 !Edrivers/gpu/drm/drm_edid.c
2252 <title>Rectangle Utilities Reference</title>
2253 !Pinclude/drm/drm_rect.h rect utils
2254 !Iinclude/drm/drm_rect.h
2255 !Edrivers/gpu/drm/drm_rect.c
2258 <title>Flip-work Helper Reference</title>
2259 !Pinclude/drm/drm_flip_work.h flip utils
2260 !Iinclude/drm/drm_flip_work.h
2261 !Edrivers/gpu/drm/drm_flip_work.c
2264 <title>HDMI Infoframes Helper Reference</title>
2266 Strictly speaking this is not a DRM helper library but generally useable
2267 by any driver interfacing with HDMI outputs like v4l or alsa drivers.
2268 But it nicely fits into the overall topic of mode setting helper
2269 libraries and hence is also included here.
2271 !Iinclude/linux/hdmi.h
2272 !Edrivers/video/hdmi.c
2276 <!-- Internals: kms properties -->
2278 <sect1 id="drm-kms-properties">
2279 <title>KMS Properties</title>
2281 Drivers may need to expose additional parameters to applications than
2282 those described in the previous sections. KMS supports attaching
2283 properties to CRTCs, connectors and planes and offers a userspace API to
2284 list, get and set the property values.
2287 Properties are identified by a name that uniquely defines the property
2288 purpose, and store an associated value. For all property types except blob
2289 properties the value is a 64-bit unsigned integer.
2292 KMS differentiates between properties and property instances. Drivers
2293 first create properties and then create and associate individual instances
2294 of those properties to objects. A property can be instantiated multiple
2295 times and associated with different objects. Values are stored in property
2296 instances, and all other property information are stored in the propery
2297 and shared between all instances of the property.
2300 Every property is created with a type that influences how the KMS core
2301 handles the property. Supported property types are
2304 <term>DRM_MODE_PROP_RANGE</term>
2305 <listitem><para>Range properties report their minimum and maximum
2306 admissible values. The KMS core verifies that values set by
2307 application fit in that range.</para></listitem>
2310 <term>DRM_MODE_PROP_ENUM</term>
2311 <listitem><para>Enumerated properties take a numerical value that
2312 ranges from 0 to the number of enumerated values defined by the
2313 property minus one, and associate a free-formed string name to each
2314 value. Applications can retrieve the list of defined value-name pairs
2315 and use the numerical value to get and set property instance values.
2319 <term>DRM_MODE_PROP_BITMASK</term>
2320 <listitem><para>Bitmask properties are enumeration properties that
2321 additionally restrict all enumerated values to the 0..63 range.
2322 Bitmask property instance values combine one or more of the
2323 enumerated bits defined by the property.</para></listitem>
2326 <term>DRM_MODE_PROP_BLOB</term>
2327 <listitem><para>Blob properties store a binary blob without any format
2328 restriction. The binary blobs are created as KMS standalone objects,
2329 and blob property instance values store the ID of their associated
2331 <para>Blob properties are only used for the connector EDID property
2332 and cannot be created by drivers.</para></listitem>
2337 To create a property drivers call one of the following functions depending
2338 on the property type. All property creation functions take property flags
2339 and name, as well as type-specific arguments.
2342 <synopsis>struct drm_property *drm_property_create_range(struct drm_device *dev, int flags,
2344 uint64_t min, uint64_t max);</synopsis>
2345 <para>Create a range property with the given minimum and maximum
2349 <synopsis>struct drm_property *drm_property_create_enum(struct drm_device *dev, int flags,
2351 const struct drm_prop_enum_list *props,
2352 int num_values);</synopsis>
2353 <para>Create an enumerated property. The <parameter>props</parameter>
2354 argument points to an array of <parameter>num_values</parameter>
2355 value-name pairs.</para>
2358 <synopsis>struct drm_property *drm_property_create_bitmask(struct drm_device *dev,
2359 int flags, const char *name,
2360 const struct drm_prop_enum_list *props,
2361 int num_values);</synopsis>
2362 <para>Create a bitmask property. The <parameter>props</parameter>
2363 argument points to an array of <parameter>num_values</parameter>
2364 value-name pairs.</para>
2369 Properties can additionally be created as immutable, in which case they
2370 will be read-only for applications but can be modified by the driver. To
2371 create an immutable property drivers must set the DRM_MODE_PROP_IMMUTABLE
2372 flag at property creation time.
2375 When no array of value-name pairs is readily available at property
2376 creation time for enumerated or range properties, drivers can create
2377 the property using the <function>drm_property_create</function> function
2378 and manually add enumeration value-name pairs by calling the
2379 <function>drm_property_add_enum</function> function. Care must be taken to
2380 properly specify the property type through the <parameter>flags</parameter>
2384 After creating properties drivers can attach property instances to CRTC,
2385 connector and plane objects by calling the
2386 <function>drm_object_attach_property</function>. The function takes a
2387 pointer to the target object, a pointer to the previously created property
2388 and an initial instance value.
2392 <!-- Internals: vertical blanking -->
2394 <sect1 id="drm-vertical-blank">
2395 <title>Vertical Blanking</title>
2397 Vertical blanking plays a major role in graphics rendering. To achieve
2398 tear-free display, users must synchronize page flips and/or rendering to
2399 vertical blanking. The DRM API offers ioctls to perform page flips
2400 synchronized to vertical blanking and wait for vertical blanking.
2403 The DRM core handles most of the vertical blanking management logic, which
2404 involves filtering out spurious interrupts, keeping race-free blanking
2405 counters, coping with counter wrap-around and resets and keeping use
2406 counts. It relies on the driver to generate vertical blanking interrupts
2407 and optionally provide a hardware vertical blanking counter. Drivers must
2408 implement the following operations.
2412 <synopsis>int (*enable_vblank) (struct drm_device *dev, int crtc);
2413 void (*disable_vblank) (struct drm_device *dev, int crtc);</synopsis>
2415 Enable or disable vertical blanking interrupts for the given CRTC.
2419 <synopsis>u32 (*get_vblank_counter) (struct drm_device *dev, int crtc);</synopsis>
2421 Retrieve the value of the vertical blanking counter for the given
2422 CRTC. If the hardware maintains a vertical blanking counter its value
2423 should be returned. Otherwise drivers can use the
2424 <function>drm_vblank_count</function> helper function to handle this
2430 Drivers must initialize the vertical blanking handling core with a call to
2431 <function>drm_vblank_init</function> in their
2432 <methodname>load</methodname> operation. The function will set the struct
2433 <structname>drm_device</structname>
2434 <structfield>vblank_disable_allowed</structfield> field to 0. This will
2435 keep vertical blanking interrupts enabled permanently until the first mode
2436 set operation, where <structfield>vblank_disable_allowed</structfield> is
2437 set to 1. The reason behind this is not clear. Drivers can set the field
2438 to 1 after <function>calling drm_vblank_init</function> to make vertical
2439 blanking interrupts dynamically managed from the beginning.
2442 Vertical blanking interrupts can be enabled by the DRM core or by drivers
2443 themselves (for instance to handle page flipping operations). The DRM core
2444 maintains a vertical blanking use count to ensure that the interrupts are
2445 not disabled while a user still needs them. To increment the use count,
2446 drivers call <function>drm_vblank_get</function>. Upon return vertical
2447 blanking interrupts are guaranteed to be enabled.
2450 To decrement the use count drivers call
2451 <function>drm_vblank_put</function>. Only when the use count drops to zero
2452 will the DRM core disable the vertical blanking interrupts after a delay
2453 by scheduling a timer. The delay is accessible through the vblankoffdelay
2454 module parameter or the <varname>drm_vblank_offdelay</varname> global
2455 variable and expressed in milliseconds. Its default value is 5000 ms.
2458 When a vertical blanking interrupt occurs drivers only need to call the
2459 <function>drm_handle_vblank</function> function to account for the
2463 Resources allocated by <function>drm_vblank_init</function> must be freed
2464 with a call to <function>drm_vblank_cleanup</function> in the driver
2465 <methodname>unload</methodname> operation handler.
2469 <!-- Internals: open/close, file operations and ioctls -->
2472 <title>Open/Close, File Operations and IOCTLs</title>
2474 <title>Open and Close</title>
2475 <synopsis>int (*firstopen) (struct drm_device *);
2476 void (*lastclose) (struct drm_device *);
2477 int (*open) (struct drm_device *, struct drm_file *);
2478 void (*preclose) (struct drm_device *, struct drm_file *);
2479 void (*postclose) (struct drm_device *, struct drm_file *);</synopsis>
2480 <abstract>Open and close handlers. None of those methods are mandatory.
2483 The <methodname>firstopen</methodname> method is called by the DRM core
2484 for legacy UMS (User Mode Setting) drivers only when an application
2485 opens a device that has no other opened file handle. UMS drivers can
2486 implement it to acquire device resources. KMS drivers can't use the
2487 method and must acquire resources in the <methodname>load</methodname>
2491 Similarly the <methodname>lastclose</methodname> method is called when
2492 the last application holding a file handle opened on the device closes
2493 it, for both UMS and KMS drivers. Additionally, the method is also
2494 called at module unload time or, for hot-pluggable devices, when the
2495 device is unplugged. The <methodname>firstopen</methodname> and
2496 <methodname>lastclose</methodname> calls can thus be unbalanced.
2499 The <methodname>open</methodname> method is called every time the device
2500 is opened by an application. Drivers can allocate per-file private data
2501 in this method and store them in the struct
2502 <structname>drm_file</structname> <structfield>driver_priv</structfield>
2503 field. Note that the <methodname>open</methodname> method is called
2504 before <methodname>firstopen</methodname>.
2507 The close operation is split into <methodname>preclose</methodname> and
2508 <methodname>postclose</methodname> methods. Drivers must stop and
2509 cleanup all per-file operations in the <methodname>preclose</methodname>
2510 method. For instance pending vertical blanking and page flip events must
2511 be cancelled. No per-file operation is allowed on the file handle after
2512 returning from the <methodname>preclose</methodname> method.
2515 Finally the <methodname>postclose</methodname> method is called as the
2516 last step of the close operation, right before calling the
2517 <methodname>lastclose</methodname> method if no other open file handle
2518 exists for the device. Drivers that have allocated per-file private data
2519 in the <methodname>open</methodname> method should free it here.
2522 The <methodname>lastclose</methodname> method should restore CRTC and
2523 plane properties to default value, so that a subsequent open of the
2524 device will not inherit state from the previous user. It can also be
2525 used to execute delayed power switching state changes, e.g. in
2526 conjunction with the vga-switcheroo infrastructure. Beyond that KMS
2527 drivers should not do any further cleanup. Only legacy UMS drivers might
2528 need to clean up device state so that the vga console or an independent
2529 fbdev driver could take over.
2533 <title>File Operations</title>
2534 <synopsis>const struct file_operations *fops</synopsis>
2535 <abstract>File operations for the DRM device node.</abstract>
2537 Drivers must define the file operations structure that forms the DRM
2538 userspace API entry point, even though most of those operations are
2539 implemented in the DRM core. The <methodname>open</methodname>,
2540 <methodname>release</methodname> and <methodname>ioctl</methodname>
2541 operations are handled by
2543 .owner = THIS_MODULE,
2545 .release = drm_release,
2546 .unlocked_ioctl = drm_ioctl,
2547 #ifdef CONFIG_COMPAT
2548 .compat_ioctl = drm_compat_ioctl,
2553 Drivers that implement private ioctls that requires 32/64bit
2554 compatibility support must provide their own
2555 <methodname>compat_ioctl</methodname> handler that processes private
2556 ioctls and calls <function>drm_compat_ioctl</function> for core ioctls.
2559 The <methodname>read</methodname> and <methodname>poll</methodname>
2560 operations provide support for reading DRM events and polling them. They
2565 .llseek = no_llseek,
2569 The memory mapping implementation varies depending on how the driver
2570 manages memory. Pre-GEM drivers will use <function>drm_mmap</function>,
2571 while GEM-aware drivers will use <function>drm_gem_mmap</function>. See
2572 <xref linkend="drm-gem"/>.
2574 .mmap = drm_gem_mmap,
2578 No other file operation is supported by the DRM API.
2582 <title>IOCTLs</title>
2583 <synopsis>struct drm_ioctl_desc *ioctls;
2584 int num_ioctls;</synopsis>
2585 <abstract>Driver-specific ioctls descriptors table.</abstract>
2587 Driver-specific ioctls numbers start at DRM_COMMAND_BASE. The ioctls
2588 descriptors table is indexed by the ioctl number offset from the base
2589 value. Drivers can use the DRM_IOCTL_DEF_DRV() macro to initialize the
2593 <programlisting>DRM_IOCTL_DEF_DRV(ioctl, func, flags)</programlisting>
2595 <parameter>ioctl</parameter> is the ioctl name. Drivers must define
2596 the DRM_##ioctl and DRM_IOCTL_##ioctl macros to the ioctl number
2597 offset from DRM_COMMAND_BASE and the ioctl number respectively. The
2598 first macro is private to the device while the second must be exposed
2599 to userspace in a public header.
2602 <parameter>func</parameter> is a pointer to the ioctl handler function
2603 compatible with the <type>drm_ioctl_t</type> type.
2604 <programlisting>typedef int drm_ioctl_t(struct drm_device *dev, void *data,
2605 struct drm_file *file_priv);</programlisting>
2608 <parameter>flags</parameter> is a bitmask combination of the following
2609 values. It restricts how the ioctl is allowed to be called.
2612 DRM_AUTH - Only authenticated callers allowed
2615 DRM_MASTER - The ioctl can only be called on the master file
2619 DRM_ROOT_ONLY - Only callers with the SYSADMIN capability allowed
2622 DRM_CONTROL_ALLOW - The ioctl can only be called on a control
2626 DRM_UNLOCKED - The ioctl handler will be called without locking
2627 the DRM global mutex
2635 <title>Legacy Support Code</title>
2637 The section very brievely covers some of the old legacy support code which
2638 is only used by old DRM drivers which have done a so-called shadow-attach
2639 to the underlying device instead of registering as a real driver. This
2640 also includes some of the old generic buffer mangement and command
2641 submission code. Do not use any of this in new and modern drivers.
2645 <title>Legacy Suspend/Resume</title>
2647 The DRM core provides some suspend/resume code, but drivers wanting full
2648 suspend/resume support should provide save() and restore() functions.
2649 These are called at suspend, hibernate, or resume time, and should perform
2650 any state save or restore required by your device across suspend or
2653 <synopsis>int (*suspend) (struct drm_device *, pm_message_t state);
2654 int (*resume) (struct drm_device *);</synopsis>
2656 Those are legacy suspend and resume methods which
2657 <emphasis>only</emphasis> work with the legacy shadow-attach driver
2658 registration functions. New driver should use the power management
2659 interface provided by their bus type (usually through
2660 the struct <structname>device_driver</structname> dev_pm_ops) and set
2661 these methods to NULL.
2666 <title>Legacy DMA Services</title>
2668 This should cover how DMA mapping etc. is supported by the core.
2669 These functions are deprecated and should not be used.
2678 - Document the struct_mutex catch-all lock
2679 - Document connector properties
2681 - Why is the load method optional?
2682 - What are drivers supposed to set the initial display state to, and how?
2683 Connector's DPMS states are not initialized and are thus equal to
2684 DRM_MODE_DPMS_ON. The fbcon compatibility layer calls
2685 drm_helper_disable_unused_functions(), which disables unused encoders and
2686 CRTCs, but doesn't touch the connectors' DPMS state, and
2687 drm_helper_connector_dpms() in reaction to fbdev blanking events. Do drivers
2688 that don't implement (or just don't use) fbcon compatibility need to call
2689 those functions themselves?
2690 - KMS drivers must call drm_vblank_pre_modeset() and drm_vblank_post_modeset()
2691 around mode setting. Should this be done in the DRM core?
2692 - vblank_disable_allowed is set to 1 in the first drm_vblank_post_modeset()
2693 call and never set back to 0. It seems to be safe to permanently set it to 1
2694 in drm_vblank_init() for KMS driver, and it might be safe for UMS drivers as
2695 well. This should be investigated.
2696 - crtc and connector .save and .restore operations are only used internally in
2697 drivers, should they be removed from the core?
2698 - encoder mid-layer .save and .restore operations are only used internally in
2699 drivers, should they be removed from the core?
2700 - encoder mid-layer .detect operation is only used internally in drivers,
2701 should it be removed from the core?
2704 <!-- External interfaces -->
2706 <chapter id="drmExternals">
2707 <title>Userland interfaces</title>
2709 The DRM core exports several interfaces to applications,
2710 generally intended to be used through corresponding libdrm
2711 wrapper functions. In addition, drivers export device-specific
2712 interfaces for use by userspace drivers & device-aware
2713 applications through ioctls and sysfs files.
2716 External interfaces include: memory mapping, context management,
2717 DMA operations, AGP management, vblank control, fence
2718 management, memory management, and output management.
2721 Cover generic ioctls and sysfs layout here. We only need high-level
2722 info, since man pages should cover the rest.
2725 <!-- External: render nodes -->
2728 <title>Render nodes</title>
2730 DRM core provides multiple character-devices for user-space to use.
2731 Depending on which device is opened, user-space can perform a different
2732 set of operations (mainly ioctls). The primary node is always created
2733 and called card<num>. Additionally, a currently
2734 unused control node, called controlD<num> is also
2735 created. The primary node provides all legacy operations and
2736 historically was the only interface used by userspace. With KMS, the
2737 control node was introduced. However, the planned KMS control interface
2738 has never been written and so the control node stays unused to date.
2741 With the increased use of offscreen renderers and GPGPU applications,
2742 clients no longer require running compositors or graphics servers to
2743 make use of a GPU. But the DRM API required unprivileged clients to
2744 authenticate to a DRM-Master prior to getting GPU access. To avoid this
2745 step and to grant clients GPU access without authenticating, render
2746 nodes were introduced. Render nodes solely serve render clients, that
2747 is, no modesetting or privileged ioctls can be issued on render nodes.
2748 Only non-global rendering commands are allowed. If a driver supports
2749 render nodes, it must advertise it via the DRIVER_RENDER
2750 DRM driver capability. If not supported, the primary node must be used
2751 for render clients together with the legacy drmAuth authentication
2755 If a driver advertises render node support, DRM core will create a
2756 separate render node called renderD<num>. There will
2757 be one render node per device. No ioctls except PRIME-related ioctls
2758 will be allowed on this node. Especially GEM_OPEN will be
2759 explicitly prohibited. Render nodes are designed to avoid the
2760 buffer-leaks, which occur if clients guess the flink names or mmap
2761 offsets on the legacy interface. Additionally to this basic interface,
2762 drivers must mark their driver-dependent render-only ioctls as
2763 DRM_RENDER_ALLOW so render clients can use them. Driver
2764 authors must be careful not to allow any privileged ioctls on render
2768 With render nodes, user-space can now control access to the render node
2769 via basic file-system access-modes. A running graphics server which
2770 authenticates clients on the privileged primary/legacy node is no longer
2771 required. Instead, a client can open the render node and is immediately
2772 granted GPU access. Communication between clients (or servers) is done
2773 via PRIME. FLINK from render node to legacy node is not supported. New
2774 clients must not use the insecure FLINK interface.
2777 Besides dropping all modeset/global ioctls, render nodes also drop the
2778 DRM-Master concept. There is no reason to associate render clients with
2779 a DRM-Master as they are independent of any graphics server. Besides,
2780 they must work without any running master, anyway.
2781 Drivers must be able to run without a master object if they support
2782 render nodes. If, on the other hand, a driver requires shared state
2783 between clients which is visible to user-space and accessible beyond
2784 open-file boundaries, they cannot support render nodes.
2788 <!-- External: vblank handling -->
2791 <title>VBlank event handling</title>
2793 The DRM core exposes two vertical blank related ioctls:
2796 <term>DRM_IOCTL_WAIT_VBLANK</term>
2799 This takes a struct drm_wait_vblank structure as its argument,
2800 and it is used to block or request a signal when a specified
2801 vblank event occurs.
2806 <term>DRM_IOCTL_MODESET_CTL</term>
2809 This should be called by application level drivers before and
2810 after mode setting, since on many devices the vertical blank
2811 counter is reset at that time. Internally, the DRM snapshots
2812 the last vblank count when the ioctl is called with the
2813 _DRM_PRE_MODESET command, so that the counter won't go backwards
2814 (which is dealt with when _DRM_POST_MODESET is used).
2819 <!--!Edrivers/char/drm/drm_irq.c-->
2825 <part id="drmDrivers">
2826 <title>DRM Drivers</title>
2830 This second part of the DRM Developer's Guide documents driver code,
2831 implementation details and also all the driver-specific userspace
2832 interfaces. Especially since all hardware-acceleration interfaces to
2833 userspace are driver specific for efficiency and other reasons these
2834 interfaces can be rather substantial. Hence every driver has its own
2839 <chapter id="drmI915">
2840 <title>drm/i915 Intel GFX Driver</title>
2842 The drm/i915 driver supports all (with the exception of some very early
2843 models) integrated GFX chipsets with both Intel display and rendering
2844 blocks. This excludes a set of SoC platforms with an SGX rendering unit,
2845 those have basic support through the gma500 drm driver.
2848 <title>Display Hardware Handling</title>
2850 This section covers everything related to the display hardware including
2851 the mode setting infrastructure, plane, sprite and cursor handling and
2852 display, output probing and related topics.
2855 <title>Mode Setting Infrastructure</title>
2857 The i915 driver is thus far the only DRM driver which doesn't use the
2858 common DRM helper code to implement mode setting sequences. Thus it
2859 has its own tailor-made infrastructure for executing a display
2860 configuration change.
2864 <title>Plane Configuration</title>
2866 This section covers plane configuration and composition with the
2867 primary plane, sprites, cursors and overlays. This includes the
2868 infrastructure to do atomic vsync'ed updates of all this state and
2869 also tightly coupled topics like watermark setup and computation,
2870 framebuffer compression and panel self refresh.
2874 <title>Output Probing</title>
2876 This section covers output probing and related infrastructure like the
2877 hotplug interrupt storm detection and mitigation code. Note that the
2878 i915 driver still uses most of the common DRM helper code for output
2879 probing, so those sections fully apply.
2885 <title>Memory Management and Command Submission</title>
2887 This sections covers all things related to the GEM implementation in the
projects / firefly-linux-kernel-4.4.55.git / blob