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>
35 <year>2008-2009</year>
37 <holder>Intel Corporation</holder>
38 <holder>Laurent Pinchart</holder>
43 The contents of this file may be used under the terms of the GNU
44 General Public License version 2 (the "GPL") as distributed in
45 the kernel source COPYING file.
50 <!-- Put document revisions here, newest first. -->
52 <revnumber>1.0</revnumber>
53 <date>2012-07-13</date>
54 <authorinitials>LP</authorinitials>
55 <revremark>Added extensive documentation about driver internals.
64 <title>DRM Core</title>
67 This first part of the DRM Developer's Guide documents core DRM code,
68 helper libraries for writting drivers and generic userspace interfaces
69 exposed by DRM drivers.
73 <chapter id="drmIntroduction">
74 <title>Introduction</title>
76 The Linux DRM layer contains code intended to support the needs
77 of complex graphics devices, usually containing programmable
78 pipelines well suited to 3D graphics acceleration. Graphics
79 drivers in the kernel may make use of DRM functions to make
80 tasks like memory management, interrupt handling and DMA easier,
81 and provide a uniform interface to applications.
84 A note on versions: this guide covers features found in the DRM
85 tree, including the TTM memory manager, output configuration and
86 mode setting, and the new vblank internals, in addition to all
87 the regular features found in current kernels.
90 [Insert diagram of typical DRM stack here]
96 <chapter id="drmInternals">
97 <title>DRM Internals</title>
99 This chapter documents DRM internals relevant to driver authors
100 and developers working to add support for the latest features to
104 First, we go over some typical driver initialization
105 requirements, like setting up command buffers, creating an
106 initial output configuration, and initializing core services.
107 Subsequent sections cover core internals in more detail,
108 providing implementation notes and examples.
111 The DRM layer provides several services to graphics drivers,
112 many of them driven by the application interfaces it provides
113 through libdrm, the library that wraps most of the DRM ioctls.
114 These include vblank event handling, memory
115 management, output management, framebuffer management, command
116 submission & fencing, suspend/resume support, and DMA
120 <!-- Internals: driver init -->
123 <title>Driver Initialization</title>
125 At the core of every DRM driver is a <structname>drm_driver</structname>
126 structure. Drivers typically statically initialize a drm_driver structure,
127 and then pass it to one of the <function>drm_*_init()</function> functions
128 to register it with the DRM subsystem.
131 The <structname>drm_driver</structname> structure contains static
132 information that describes the driver and features it supports, and
133 pointers to methods that the DRM core will call to implement the DRM API.
134 We will first go through the <structname>drm_driver</structname> static
135 information fields, and will then describe individual operations in
136 details as they get used in later sections.
139 <title>Driver Information</title>
141 <title>Driver Features</title>
143 Drivers inform the DRM core about their requirements and supported
144 features by setting appropriate flags in the
145 <structfield>driver_features</structfield> field. Since those flags
146 influence the DRM core behaviour since registration time, most of them
147 must be set to registering the <structname>drm_driver</structname>
150 <synopsis>u32 driver_features;</synopsis>
152 <title>Driver Feature Flags</title>
154 <term>DRIVER_USE_AGP</term>
156 Driver uses AGP interface, the DRM core will manage AGP resources.
160 <term>DRIVER_REQUIRE_AGP</term>
162 Driver needs AGP interface to function. AGP initialization failure
163 will become a fatal error.
167 <term>DRIVER_PCI_DMA</term>
169 Driver is capable of PCI DMA, mapping of PCI DMA buffers to
170 userspace will be enabled. Deprecated.
174 <term>DRIVER_SG</term>
176 Driver can perform scatter/gather DMA, allocation and mapping of
177 scatter/gather buffers will be enabled. Deprecated.
181 <term>DRIVER_HAVE_DMA</term>
183 Driver supports DMA, the userspace DMA API will be supported.
188 <term>DRIVER_HAVE_IRQ</term><term>DRIVER_IRQ_SHARED</term>
190 DRIVER_HAVE_IRQ indicates whether the driver has an IRQ handler
191 managed by the DRM Core. The core will support simple IRQ handler
192 installation when the flag is set. The installation process is
193 described in <xref linkend="drm-irq-registration"/>.</para>
194 <para>DRIVER_IRQ_SHARED indicates whether the device & handler
195 support shared IRQs (note that this is required of PCI drivers).
199 <term>DRIVER_GEM</term>
201 Driver use the GEM memory manager.
205 <term>DRIVER_MODESET</term>
207 Driver supports mode setting interfaces (KMS).
211 <term>DRIVER_PRIME</term>
213 Driver implements DRM PRIME buffer sharing.
217 <term>DRIVER_RENDER</term>
219 Driver supports dedicated render nodes.
225 <title>Major, Minor and Patchlevel</title>
228 int patchlevel;</synopsis>
230 The DRM core identifies driver versions by a major, minor and patch
231 level triplet. The information is printed to the kernel log at
232 initialization time and passed to userspace through the
233 DRM_IOCTL_VERSION ioctl.
236 The major and minor numbers are also used to verify the requested driver
237 API version passed to DRM_IOCTL_SET_VERSION. When the driver API changes
238 between minor versions, applications can call DRM_IOCTL_SET_VERSION to
239 select a specific version of the API. If the requested major isn't equal
240 to the driver major, or the requested minor is larger than the driver
241 minor, the DRM_IOCTL_SET_VERSION call will return an error. Otherwise
242 the driver's set_version() method will be called with the requested
247 <title>Name, Description and Date</title>
248 <synopsis>char *name;
250 char *date;</synopsis>
252 The driver name is printed to the kernel log at initialization time,
253 used for IRQ registration and passed to userspace through
257 The driver description is a purely informative string passed to
258 userspace through the DRM_IOCTL_VERSION ioctl and otherwise unused by
262 The driver date, formatted as YYYYMMDD, is meant to identify the date of
263 the latest modification to the driver. However, as most drivers fail to
264 update it, its value is mostly useless. The DRM core prints it to the
265 kernel log at initialization time and passes it to userspace through the
266 DRM_IOCTL_VERSION ioctl.
271 <title>Driver Load</title>
273 The <methodname>load</methodname> method is the driver and device
274 initialization entry point. The method is responsible for allocating and
275 initializing driver private data, performing resource allocation and
276 mapping (e.g. acquiring
277 clocks, mapping registers or allocating command buffers), initializing
278 the memory manager (<xref linkend="drm-memory-management"/>), installing
279 the IRQ handler (<xref linkend="drm-irq-registration"/>), setting up
280 vertical blanking handling (<xref linkend="drm-vertical-blank"/>), mode
281 setting (<xref linkend="drm-mode-setting"/>) and initial output
282 configuration (<xref linkend="drm-kms-init"/>).
285 If compatibility is a concern (e.g. with drivers converted over from
286 User Mode Setting to Kernel Mode Setting), care must be taken to prevent
287 device initialization and control that is incompatible with currently
288 active userspace drivers. For instance, if user level mode setting
289 drivers are in use, it would be problematic to perform output discovery
290 & configuration at load time. Likewise, if user-level drivers
291 unaware of memory management are in use, memory management and command
292 buffer setup may need to be omitted. These requirements are
293 driver-specific, and care needs to be taken to keep both old and new
294 applications and libraries working.
296 <synopsis>int (*load) (struct drm_device *, unsigned long flags);</synopsis>
298 The method takes two arguments, a pointer to the newly created
299 <structname>drm_device</structname> and flags. The flags are used to
300 pass the <structfield>driver_data</structfield> field of the device id
301 corresponding to the device passed to <function>drm_*_init()</function>.
302 Only PCI devices currently use this, USB and platform DRM drivers have
303 their <methodname>load</methodname> method called with flags to 0.
306 <title>Driver Private Data</title>
308 The driver private hangs off the main
309 <structname>drm_device</structname> structure and can be used for
310 tracking various device-specific bits of information, like register
311 offsets, command buffer status, register state for suspend/resume, etc.
312 At load time, a driver may simply allocate one and set
313 <structname>drm_device</structname>.<structfield>dev_priv</structfield>
314 appropriately; it should be freed and
315 <structname>drm_device</structname>.<structfield>dev_priv</structfield>
316 set to NULL when the driver is unloaded.
319 <sect3 id="drm-irq-registration">
320 <title>IRQ Registration</title>
322 The DRM core tries to facilitate IRQ handler registration and
323 unregistration by providing <function>drm_irq_install</function> and
324 <function>drm_irq_uninstall</function> functions. Those functions only
325 support a single interrupt per device, devices that use more than one
326 IRQs need to be handled manually.
329 <title>Managed IRQ Registration</title>
331 Both the <function>drm_irq_install</function> and
332 <function>drm_irq_uninstall</function> functions get the device IRQ by
333 calling <function>drm_dev_to_irq</function>. This inline function will
334 call a bus-specific operation to retrieve the IRQ number. For platform
335 devices, <function>platform_get_irq</function>(..., 0) is used to
336 retrieve the IRQ number.
339 <function>drm_irq_install</function> starts by calling the
340 <methodname>irq_preinstall</methodname> driver operation. The operation
341 is optional and must make sure that the interrupt will not get fired by
342 clearing all pending interrupt flags or disabling the interrupt.
345 The IRQ will then be requested by a call to
346 <function>request_irq</function>. If the DRIVER_IRQ_SHARED driver
347 feature flag is set, a shared (IRQF_SHARED) IRQ handler will be
351 The IRQ handler function must be provided as the mandatory irq_handler
352 driver operation. It will get passed directly to
353 <function>request_irq</function> and thus has the same prototype as all
354 IRQ handlers. It will get called with a pointer to the DRM device as the
358 Finally the function calls the optional
359 <methodname>irq_postinstall</methodname> driver operation. The operation
360 usually enables interrupts (excluding the vblank interrupt, which is
361 enabled separately), but drivers may choose to enable/disable interrupts
365 <function>drm_irq_uninstall</function> is similarly used to uninstall an
366 IRQ handler. It starts by waking up all processes waiting on a vblank
367 interrupt to make sure they don't hang, and then calls the optional
368 <methodname>irq_uninstall</methodname> driver operation. The operation
369 must disable all hardware interrupts. Finally the function frees the IRQ
370 by calling <function>free_irq</function>.
374 <title>Manual IRQ Registration</title>
376 Drivers that require multiple interrupt handlers can't use the managed
377 IRQ registration functions. In that case IRQs must be registered and
378 unregistered manually (usually with the <function>request_irq</function>
379 and <function>free_irq</function> functions, or their devm_* equivalent).
382 When manually registering IRQs, drivers must not set the DRIVER_HAVE_IRQ
383 driver feature flag, and must not provide the
384 <methodname>irq_handler</methodname> driver operation. They must set the
385 <structname>drm_device</structname> <structfield>irq_enabled</structfield>
386 field to 1 upon registration of the IRQs, and clear it to 0 after
387 unregistering the IRQs.
392 <title>Memory Manager Initialization</title>
394 Every DRM driver requires a memory manager which must be initialized at
395 load time. DRM currently contains two memory managers, the Translation
396 Table Manager (TTM) and the Graphics Execution Manager (GEM).
397 This document describes the use of the GEM memory manager only. See
398 <xref linkend="drm-memory-management"/> for details.
402 <title>Miscellaneous Device Configuration</title>
404 Another task that may be necessary for PCI devices during configuration
405 is mapping the video BIOS. On many devices, the VBIOS describes device
406 configuration, LCD panel timings (if any), and contains flags indicating
407 device state. Mapping the BIOS can be done using the pci_map_rom() call,
408 a convenience function that takes care of mapping the actual ROM,
409 whether it has been shadowed into memory (typically at address 0xc0000)
410 or exists on the PCI device in the ROM BAR. Note that after the ROM has
411 been mapped and any necessary information has been extracted, it should
412 be unmapped; on many devices, the ROM address decoder is shared with
413 other BARs, so leaving it mapped could cause undesired behaviour like
414 hangs or memory corruption.
415 <!--!Fdrivers/pci/rom.c pci_map_rom-->
421 <!-- Internals: memory management -->
423 <sect1 id="drm-memory-management">
424 <title>Memory management</title>
426 Modern Linux systems require large amount of graphics memory to store
427 frame buffers, textures, vertices and other graphics-related data. Given
428 the very dynamic nature of many of that data, managing graphics memory
429 efficiently is thus crucial for the graphics stack and plays a central
430 role in the DRM infrastructure.
433 The DRM core includes two memory managers, namely Translation Table Maps
434 (TTM) and Graphics Execution Manager (GEM). TTM was the first DRM memory
435 manager to be developed and tried to be a one-size-fits-them all
436 solution. It provides a single userspace API to accommodate the need of
437 all hardware, supporting both Unified Memory Architecture (UMA) devices
438 and devices with dedicated video RAM (i.e. most discrete video cards).
439 This resulted in a large, complex piece of code that turned out to be
440 hard to use for driver development.
443 GEM started as an Intel-sponsored project in reaction to TTM's
444 complexity. Its design philosophy is completely different: instead of
445 providing a solution to every graphics memory-related problems, GEM
446 identified common code between drivers and created a support library to
447 share it. GEM has simpler initialization and execution requirements than
448 TTM, but has no video RAM management capabitilies and is thus limited to
452 <title>The Translation Table Manager (TTM)</title>
454 TTM design background and information belongs here.
457 <title>TTM initialization</title>
458 <warning><para>This section is outdated.</para></warning>
460 Drivers wishing to support TTM must fill out a drm_bo_driver
461 structure. The structure contains several fields with function
462 pointers for initializing the TTM, allocating and freeing memory,
463 waiting for command completion and fence synchronization, and memory
464 migration. See the radeon_ttm.c file for an example of usage.
467 The ttm_global_reference structure is made up of several fields:
470 struct ttm_global_reference {
471 enum ttm_global_types global_type;
474 int (*init) (struct ttm_global_reference *);
475 void (*release) (struct ttm_global_reference *);
479 There should be one global reference structure for your memory
480 manager as a whole, and there will be others for each object
481 created by the memory manager at runtime. Your global TTM should
482 have a type of TTM_GLOBAL_TTM_MEM. The size field for the global
483 object should be sizeof(struct ttm_mem_global), and the init and
484 release hooks should point at your driver-specific init and
485 release routines, which probably eventually call
486 ttm_mem_global_init and ttm_mem_global_release, respectively.
489 Once your global TTM accounting structure is set up and initialized
490 by calling ttm_global_item_ref() on it,
491 you need to create a buffer object TTM to
492 provide a pool for buffer object allocation by clients and the
493 kernel itself. The type of this object should be TTM_GLOBAL_TTM_BO,
494 and its size should be sizeof(struct ttm_bo_global). Again,
495 driver-specific init and release functions may be provided,
496 likely eventually calling ttm_bo_global_init() and
497 ttm_bo_global_release(), respectively. Also, like the previous
498 object, ttm_global_item_ref() is used to create an initial reference
499 count for the TTM, which will call your initialization function.
504 <title>The Graphics Execution Manager (GEM)</title>
506 The GEM design approach has resulted in a memory manager that doesn't
507 provide full coverage of all (or even all common) use cases in its
508 userspace or kernel API. GEM exposes a set of standard memory-related
509 operations to userspace and a set of helper functions to drivers, and let
510 drivers implement hardware-specific operations with their own private API.
513 The GEM userspace API is described in the
514 <ulink url="http://lwn.net/Articles/283798/"><citetitle>GEM - the Graphics
515 Execution Manager</citetitle></ulink> article on LWN. While slightly
516 outdated, the document provides a good overview of the GEM API principles.
517 Buffer allocation and read and write operations, described as part of the
518 common GEM API, are currently implemented using driver-specific ioctls.
521 GEM is data-agnostic. It manages abstract buffer objects without knowing
522 what individual buffers contain. APIs that require knowledge of buffer
523 contents or purpose, such as buffer allocation or synchronization
524 primitives, are thus outside of the scope of GEM and must be implemented
525 using driver-specific ioctls.
528 On a fundamental level, GEM involves several operations:
530 <listitem>Memory allocation and freeing</listitem>
531 <listitem>Command execution</listitem>
532 <listitem>Aperture management at command execution time</listitem>
534 Buffer object allocation is relatively straightforward and largely
535 provided by Linux's shmem layer, which provides memory to back each
539 Device-specific operations, such as command execution, pinning, buffer
540 read & write, mapping, and domain ownership transfers are left to
541 driver-specific ioctls.
544 <title>GEM Initialization</title>
546 Drivers that use GEM must set the DRIVER_GEM bit in the struct
547 <structname>drm_driver</structname>
548 <structfield>driver_features</structfield> field. The DRM core will
549 then automatically initialize the GEM core before calling the
550 <methodname>load</methodname> operation. Behind the scene, this will
551 create a DRM Memory Manager object which provides an address space
552 pool for object allocation.
555 In a KMS configuration, drivers need to allocate and initialize a
556 command ring buffer following core GEM initialization if required by
557 the hardware. UMA devices usually have what is called a "stolen"
558 memory region, which provides space for the initial framebuffer and
559 large, contiguous memory regions required by the device. This space is
560 typically not managed by GEM, and must be initialized separately into
561 its own DRM MM object.
565 <title>GEM Objects Creation</title>
567 GEM splits creation of GEM objects and allocation of the memory that
568 backs them in two distinct operations.
571 GEM objects are represented by an instance of struct
572 <structname>drm_gem_object</structname>. Drivers usually need to extend
573 GEM objects with private information and thus create a driver-specific
574 GEM object structure type that embeds an instance of struct
575 <structname>drm_gem_object</structname>.
578 To create a GEM object, a driver allocates memory for an instance of its
579 specific GEM object type and initializes the embedded struct
580 <structname>drm_gem_object</structname> with a call to
581 <function>drm_gem_object_init</function>. The function takes a pointer to
582 the DRM device, a pointer to the GEM object and the buffer object size
586 GEM uses shmem to allocate anonymous pageable memory.
587 <function>drm_gem_object_init</function> will create an shmfs file of
588 the requested size and store it into the struct
589 <structname>drm_gem_object</structname> <structfield>filp</structfield>
590 field. The memory is used as either main storage for the object when the
591 graphics hardware uses system memory directly or as a backing store
595 Drivers are responsible for the actual physical pages allocation by
596 calling <function>shmem_read_mapping_page_gfp</function> for each page.
597 Note that they can decide to allocate pages when initializing the GEM
598 object, or to delay allocation until the memory is needed (for instance
599 when a page fault occurs as a result of a userspace memory access or
600 when the driver needs to start a DMA transfer involving the memory).
603 Anonymous pageable memory allocation is not always desired, for instance
604 when the hardware requires physically contiguous system memory as is
605 often the case in embedded devices. Drivers can create GEM objects with
606 no shmfs backing (called private GEM objects) by initializing them with
607 a call to <function>drm_gem_private_object_init</function> instead of
608 <function>drm_gem_object_init</function>. Storage for private GEM
609 objects must be managed by drivers.
612 Drivers that do not need to extend GEM objects with private information
613 can call the <function>drm_gem_object_alloc</function> function to
614 allocate and initialize a struct <structname>drm_gem_object</structname>
615 instance. The GEM core will call the optional driver
616 <methodname>gem_init_object</methodname> operation after initializing
617 the GEM object with <function>drm_gem_object_init</function>.
618 <synopsis>int (*gem_init_object) (struct drm_gem_object *obj);</synopsis>
621 No alloc-and-init function exists for private GEM objects.
625 <title>GEM Objects Lifetime</title>
627 All GEM objects are reference-counted by the GEM core. References can be
628 acquired and release by <function>calling drm_gem_object_reference</function>
629 and <function>drm_gem_object_unreference</function> respectively. The
630 caller must hold the <structname>drm_device</structname>
631 <structfield>struct_mutex</structfield> lock. As a convenience, GEM
632 provides the <function>drm_gem_object_reference_unlocked</function> and
633 <function>drm_gem_object_unreference_unlocked</function> functions that
634 can be called without holding the lock.
637 When the last reference to a GEM object is released the GEM core calls
638 the <structname>drm_driver</structname>
639 <methodname>gem_free_object</methodname> operation. That operation is
640 mandatory for GEM-enabled drivers and must free the GEM object and all
641 associated resources.
644 <synopsis>void (*gem_free_object) (struct drm_gem_object *obj);</synopsis>
645 Drivers are responsible for freeing all GEM object resources, including
646 the resources created by the GEM core. If an mmap offset has been
647 created for the object (in which case
648 <structname>drm_gem_object</structname>::<structfield>map_list</structfield>::<structfield>map</structfield>
649 is not NULL) it must be freed by a call to
650 <function>drm_gem_free_mmap_offset</function>. The shmfs backing store
651 must be released by calling <function>drm_gem_object_release</function>
652 (that function can safely be called if no shmfs backing store has been
657 <title>GEM Objects Naming</title>
659 Communication between userspace and the kernel refers to GEM objects
660 using local handles, global names or, more recently, file descriptors.
661 All of those are 32-bit integer values; the usual Linux kernel limits
662 apply to the file descriptors.
665 GEM handles are local to a DRM file. Applications get a handle to a GEM
666 object through a driver-specific ioctl, and can use that handle to refer
667 to the GEM object in other standard or driver-specific ioctls. Closing a
668 DRM file handle frees all its GEM handles and dereferences the
669 associated GEM objects.
672 To create a handle for a GEM object drivers call
673 <function>drm_gem_handle_create</function>. The function takes a pointer
674 to the DRM file and the GEM object and returns a locally unique handle.
675 When the handle is no longer needed drivers delete it with a call to
676 <function>drm_gem_handle_delete</function>. Finally the GEM object
677 associated with a handle can be retrieved by a call to
678 <function>drm_gem_object_lookup</function>.
681 Handles don't take ownership of GEM objects, they only take a reference
682 to the object that will be dropped when the handle is destroyed. To
683 avoid leaking GEM objects, drivers must make sure they drop the
684 reference(s) they own (such as the initial reference taken at object
685 creation time) as appropriate, without any special consideration for the
686 handle. For example, in the particular case of combined GEM object and
687 handle creation in the implementation of the
688 <methodname>dumb_create</methodname> operation, drivers must drop the
689 initial reference to the GEM object before returning the handle.
692 GEM names are similar in purpose to handles but are not local to DRM
693 files. They can be passed between processes to reference a GEM object
694 globally. Names can't be used directly to refer to objects in the DRM
695 API, applications must convert handles to names and names to handles
696 using the DRM_IOCTL_GEM_FLINK and DRM_IOCTL_GEM_OPEN ioctls
697 respectively. The conversion is handled by the DRM core without any
698 driver-specific support.
701 GEM also supports buffer sharing with dma-buf file descriptors through
702 PRIME. GEM-based drivers must use the provided helpers functions to
703 implement the exporting and importing correctly. See <xref linkend="drm-prime-support" />.
704 Since sharing file descriptors is inherently more secure than the
705 easily guessable and global GEM names it is the preferred buffer
706 sharing mechanism. Sharing buffers through GEM names is only supported
707 for legacy userspace. Furthermore PRIME also allows cross-device
708 buffer sharing since it is based on dma-bufs.
711 <sect3 id="drm-gem-objects-mapping">
712 <title>GEM Objects Mapping</title>
714 Because mapping operations are fairly heavyweight GEM favours
715 read/write-like access to buffers, implemented through driver-specific
716 ioctls, over mapping buffers to userspace. However, when random access
717 to the buffer is needed (to perform software rendering for instance),
718 direct access to the object can be more efficient.
721 The mmap system call can't be used directly to map GEM objects, as they
722 don't have their own file handle. Two alternative methods currently
723 co-exist to map GEM objects to userspace. The first method uses a
724 driver-specific ioctl to perform the mapping operation, calling
725 <function>do_mmap</function> under the hood. This is often considered
726 dubious, seems to be discouraged for new GEM-enabled drivers, and will
727 thus not be described here.
730 The second method uses the mmap system call on the DRM file handle.
731 <synopsis>void *mmap(void *addr, size_t length, int prot, int flags, int fd,
732 off_t offset);</synopsis>
733 DRM identifies the GEM object to be mapped by a fake offset passed
734 through the mmap offset argument. Prior to being mapped, a GEM object
735 must thus be associated with a fake offset. To do so, drivers must call
736 <function>drm_gem_create_mmap_offset</function> on the object. The
737 function allocates a fake offset range from a pool and stores the
738 offset divided by PAGE_SIZE in
739 <literal>obj->map_list.hash.key</literal>. Care must be taken not to
740 call <function>drm_gem_create_mmap_offset</function> if a fake offset
741 has already been allocated for the object. This can be tested by
742 <literal>obj->map_list.map</literal> being non-NULL.
745 Once allocated, the fake offset value
746 (<literal>obj->map_list.hash.key << PAGE_SHIFT</literal>)
747 must be passed to the application in a driver-specific way and can then
748 be used as the mmap offset argument.
751 The GEM core provides a helper method <function>drm_gem_mmap</function>
752 to handle object mapping. The method can be set directly as the mmap
753 file operation handler. It will look up the GEM object based on the
754 offset value and set the VMA operations to the
755 <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
756 field. Note that <function>drm_gem_mmap</function> doesn't map memory to
757 userspace, but relies on the driver-provided fault handler to map pages
761 To use <function>drm_gem_mmap</function>, drivers must fill the struct
762 <structname>drm_driver</structname> <structfield>gem_vm_ops</structfield>
763 field with a pointer to VM operations.
766 <synopsis>struct vm_operations_struct *gem_vm_ops
768 struct vm_operations_struct {
769 void (*open)(struct vm_area_struct * area);
770 void (*close)(struct vm_area_struct * area);
771 int (*fault)(struct vm_area_struct *vma, struct vm_fault *vmf);
775 The <methodname>open</methodname> and <methodname>close</methodname>
776 operations must update the GEM object reference count. Drivers can use
777 the <function>drm_gem_vm_open</function> and
778 <function>drm_gem_vm_close</function> helper functions directly as open
782 The fault operation handler is responsible for mapping individual pages
783 to userspace when a page fault occurs. Depending on the memory
784 allocation scheme, drivers can allocate pages at fault time, or can
785 decide to allocate memory for the GEM object at the time the object is
789 Drivers that want to map the GEM object upfront instead of handling page
790 faults can implement their own mmap file operation handler.
794 <title>Memory Coherency</title>
796 When mapped to the device or used in a command buffer, backing pages
797 for an object are flushed to memory and marked write combined so as to
798 be coherent with the GPU. Likewise, if the CPU accesses an object
799 after the GPU has finished rendering to the object, then the object
800 must be made coherent with the CPU's view of memory, usually involving
801 GPU cache flushing of various kinds. This core CPU<->GPU
802 coherency management is provided by a device-specific ioctl, which
803 evaluates an object's current domain and performs any necessary
804 flushing or synchronization to put the object into the desired
805 coherency domain (note that the object may be busy, i.e. an active
806 render target; in that case, setting the domain blocks the client and
807 waits for rendering to complete before performing any necessary
808 flushing operations).
812 <title>Command Execution</title>
814 Perhaps the most important GEM function for GPU devices is providing a
815 command execution interface to clients. Client programs construct
816 command buffers containing references to previously allocated memory
817 objects, and then submit them to GEM. At that point, GEM takes care to
818 bind all the objects into the GTT, execute the buffer, and provide
819 necessary synchronization between clients accessing the same buffers.
820 This often involves evicting some objects from the GTT and re-binding
821 others (a fairly expensive operation), and providing relocation
822 support which hides fixed GTT offsets from clients. Clients must take
823 care not to submit command buffers that reference more objects than
824 can fit in the GTT; otherwise, GEM will reject them and no rendering
825 will occur. Similarly, if several objects in the buffer require fence
826 registers to be allocated for correct rendering (e.g. 2D blits on
827 pre-965 chips), care must be taken not to require more fence registers
828 than are available to the client. Such resource management should be
829 abstracted from the client in libdrm.
833 <title>GEM Function Reference</title>
834 !Edrivers/gpu/drm/drm_gem.c
838 <title>VMA Offset Manager</title>
839 !Pdrivers/gpu/drm/drm_vma_manager.c vma offset manager
840 !Edrivers/gpu/drm/drm_vma_manager.c
841 !Iinclude/drm/drm_vma_manager.h
843 <sect2 id="drm-prime-support">
844 <title>PRIME Buffer Sharing</title>
846 PRIME is the cross device buffer sharing framework in drm, originally
847 created for the OPTIMUS range of multi-gpu platforms. To userspace
848 PRIME buffers are dma-buf based file descriptors.
851 <title>Overview and Driver Interface</title>
853 Similar to GEM global names, PRIME file descriptors are
854 also used to share buffer objects across processes. They offer
855 additional security: as file descriptors must be explicitly sent over
856 UNIX domain sockets to be shared between applications, they can't be
857 guessed like the globally unique GEM names.
860 Drivers that support the PRIME
861 API must set the DRIVER_PRIME bit in the struct
862 <structname>drm_driver</structname>
863 <structfield>driver_features</structfield> field, and implement the
864 <methodname>prime_handle_to_fd</methodname> and
865 <methodname>prime_fd_to_handle</methodname> operations.
868 <synopsis>int (*prime_handle_to_fd)(struct drm_device *dev,
869 struct drm_file *file_priv, uint32_t handle,
870 uint32_t flags, int *prime_fd);
871 int (*prime_fd_to_handle)(struct drm_device *dev,
872 struct drm_file *file_priv, int prime_fd,
873 uint32_t *handle);</synopsis>
874 Those two operations convert a handle to a PRIME file descriptor and
875 vice versa. Drivers must use the kernel dma-buf buffer sharing framework
876 to manage the PRIME file descriptors. Similar to the mode setting
877 API PRIME is agnostic to the underlying buffer object manager, as
878 long as handles are 32bit unsinged integers.
881 While non-GEM drivers must implement the operations themselves, GEM
882 drivers must use the <function>drm_gem_prime_handle_to_fd</function>
883 and <function>drm_gem_prime_fd_to_handle</function> helper functions.
884 Those helpers rely on the driver
885 <methodname>gem_prime_export</methodname> and
886 <methodname>gem_prime_import</methodname> operations to create a dma-buf
887 instance from a GEM object (dma-buf exporter role) and to create a GEM
888 object from a dma-buf instance (dma-buf importer role).
891 <synopsis>struct dma_buf * (*gem_prime_export)(struct drm_device *dev,
892 struct drm_gem_object *obj,
894 struct drm_gem_object * (*gem_prime_import)(struct drm_device *dev,
895 struct dma_buf *dma_buf);</synopsis>
896 These two operations are mandatory for GEM drivers that support
901 <title>PRIME Helper Functions</title>
902 !Pdrivers/gpu/drm/drm_prime.c PRIME Helpers
906 <title>PRIME Function References</title>
907 !Edrivers/gpu/drm/drm_prime.c
911 <!-- Internals: mode setting -->
913 <sect1 id="drm-mode-setting">
914 <title>Mode Setting</title>
916 Drivers must initialize the mode setting core by calling
917 <function>drm_mode_config_init</function> on the DRM device. The function
918 initializes the <structname>drm_device</structname>
919 <structfield>mode_config</structfield> field and never fails. Once done,
920 mode configuration must be setup by initializing the following fields.
924 <synopsis>int min_width, min_height;
925 int max_width, max_height;</synopsis>
927 Minimum and maximum width and height of the frame buffers in pixel
932 <synopsis>struct drm_mode_config_funcs *funcs;</synopsis>
933 <para>Mode setting functions.</para>
937 <title>Frame Buffer Creation</title>
938 <synopsis>struct drm_framebuffer *(*fb_create)(struct drm_device *dev,
939 struct drm_file *file_priv,
940 struct drm_mode_fb_cmd2 *mode_cmd);</synopsis>
942 Frame buffers are abstract memory objects that provide a source of
943 pixels to scanout to a CRTC. Applications explicitly request the
944 creation of frame buffers through the DRM_IOCTL_MODE_ADDFB(2) ioctls and
945 receive an opaque handle that can be passed to the KMS CRTC control,
946 plane configuration and page flip functions.
949 Frame buffers rely on the underneath memory manager for low-level memory
950 operations. When creating a frame buffer applications pass a memory
951 handle (or a list of memory handles for multi-planar formats) through
952 the <parameter>drm_mode_fb_cmd2</parameter> argument. For drivers using
953 GEM as their userspace buffer management interface this would be a GEM
954 handle. Drivers are however free to use their own backing storage object
955 handles, e.g. vmwgfx directly exposes special TTM handles to userspace
956 and so expects TTM handles in the create ioctl and not GEM handles.
959 Drivers must first validate the requested frame buffer parameters passed
960 through the mode_cmd argument. In particular this is where invalid
961 sizes, pixel formats or pitches can be caught.
964 If the parameters are deemed valid, drivers then create, initialize and
965 return an instance of struct <structname>drm_framebuffer</structname>.
966 If desired the instance can be embedded in a larger driver-specific
967 structure. Drivers must fill its <structfield>width</structfield>,
968 <structfield>height</structfield>, <structfield>pitches</structfield>,
969 <structfield>offsets</structfield>, <structfield>depth</structfield>,
970 <structfield>bits_per_pixel</structfield> and
971 <structfield>pixel_format</structfield> fields from the values passed
972 through the <parameter>drm_mode_fb_cmd2</parameter> argument. They
973 should call the <function>drm_helper_mode_fill_fb_struct</function>
974 helper function to do so.
978 The initialization of the new framebuffer instance is finalized with a
979 call to <function>drm_framebuffer_init</function> which takes a pointer
980 to DRM frame buffer operations (struct
981 <structname>drm_framebuffer_funcs</structname>). Note that this function
982 publishes the framebuffer and so from this point on it can be accessed
983 concurrently from other threads. Hence it must be the last step in the
984 driver's framebuffer initialization sequence. Frame buffer operations
988 <synopsis>int (*create_handle)(struct drm_framebuffer *fb,
989 struct drm_file *file_priv, unsigned int *handle);</synopsis>
991 Create a handle to the frame buffer underlying memory object. If
992 the frame buffer uses a multi-plane format, the handle will
993 reference the memory object associated with the first plane.
996 Drivers call <function>drm_gem_handle_create</function> to create
1001 <synopsis>void (*destroy)(struct drm_framebuffer *framebuffer);</synopsis>
1003 Destroy the frame buffer object and frees all associated
1004 resources. Drivers must call
1005 <function>drm_framebuffer_cleanup</function> to free resources
1006 allocated by the DRM core for the frame buffer object, and must
1007 make sure to unreference all memory objects associated with the
1008 frame buffer. Handles created by the
1009 <methodname>create_handle</methodname> operation are released by
1014 <synopsis>int (*dirty)(struct drm_framebuffer *framebuffer,
1015 struct drm_file *file_priv, unsigned flags, unsigned color,
1016 struct drm_clip_rect *clips, unsigned num_clips);</synopsis>
1018 This optional operation notifies the driver that a region of the
1019 frame buffer has changed in response to a DRM_IOCTL_MODE_DIRTYFB
1026 The lifetime of a drm framebuffer is controlled with a reference count,
1027 drivers can grab additional references with
1028 <function>drm_framebuffer_reference</function> </para> and drop them
1029 again with <function>drm_framebuffer_unreference</function>. For
1030 driver-private framebuffers for which the last reference is never
1031 dropped (e.g. for the fbdev framebuffer when the struct
1032 <structname>drm_framebuffer</structname> is embedded into the fbdev
1033 helper struct) drivers can manually clean up a framebuffer at module
1035 <function>drm_framebuffer_unregister_private</function>.
1038 <title>Dumb Buffer Objects</title>
1040 The KMS API doesn't standardize backing storage object creation and
1041 leaves it to driver-specific ioctls. Furthermore actually creating a
1042 buffer object even for GEM-based drivers is done through a
1043 driver-specific ioctl - GEM only has a common userspace interface for
1044 sharing and destroying objects. While not an issue for full-fledged
1045 graphics stacks that include device-specific userspace components (in
1046 libdrm for instance), this limit makes DRM-based early boot graphics
1047 unnecessarily complex.
1050 Dumb objects partly alleviate the problem by providing a standard
1051 API to create dumb buffers suitable for scanout, which can then be used
1052 to create KMS frame buffers.
1055 To support dumb objects drivers must implement the
1056 <methodname>dumb_create</methodname>,
1057 <methodname>dumb_destroy</methodname> and
1058 <methodname>dumb_map_offset</methodname> operations.
1062 <synopsis>int (*dumb_create)(struct drm_file *file_priv, struct drm_device *dev,
1063 struct drm_mode_create_dumb *args);</synopsis>
1065 The <methodname>dumb_create</methodname> operation creates a driver
1066 object (GEM or TTM handle) suitable for scanout based on the
1067 width, height and depth from the struct
1068 <structname>drm_mode_create_dumb</structname> argument. It fills the
1069 argument's <structfield>handle</structfield>,
1070 <structfield>pitch</structfield> and <structfield>size</structfield>
1071 fields with a handle for the newly created object and its line
1072 pitch and size in bytes.
1076 <synopsis>int (*dumb_destroy)(struct drm_file *file_priv, struct drm_device *dev,
1077 uint32_t handle);</synopsis>
1079 The <methodname>dumb_destroy</methodname> operation destroys a dumb
1080 object created by <methodname>dumb_create</methodname>.
1084 <synopsis>int (*dumb_map_offset)(struct drm_file *file_priv, struct drm_device *dev,
1085 uint32_t handle, uint64_t *offset);</synopsis>
1087 The <methodname>dumb_map_offset</methodname> operation associates an
1088 mmap fake offset with the object given by the handle and returns
1089 it. Drivers must use the
1090 <function>drm_gem_create_mmap_offset</function> function to
1091 associate the fake offset as described in
1092 <xref linkend="drm-gem-objects-mapping"/>.
1097 Note that dumb objects may not be used for gpu acceleration, as has been
1098 attempted on some ARM embedded platforms. Such drivers really must have
1099 a hardware-specific ioctl to allocate suitable buffer objects.
1103 <title>Output Polling</title>
1104 <synopsis>void (*output_poll_changed)(struct drm_device *dev);</synopsis>
1106 This operation notifies the driver that the status of one or more
1107 connectors has changed. Drivers that use the fb helper can just call the
1108 <function>drm_fb_helper_hotplug_event</function> function to handle this
1113 <title>Locking</title>
1115 Beside some lookup structures with their own locking (which is hidden
1116 behind the interface functions) most of the modeset state is protected
1117 by the <code>dev-<mode_config.lock</code> mutex and additionally
1118 per-crtc locks to allow cursor updates, pageflips and similar operations
1119 to occur concurrently with background tasks like output detection.
1120 Operations which cross domains like a full modeset always grab all
1121 locks. Drivers there need to protect resources shared between crtcs with
1122 additional locking. They also need to be careful to always grab the
1123 relevant crtc locks if a modset functions touches crtc state, e.g. for
1124 load detection (which does only grab the <code>mode_config.lock</code>
1125 to allow concurrent screen updates on live crtcs).
1130 <!-- Internals: kms initialization and cleanup -->
1132 <sect1 id="drm-kms-init">
1133 <title>KMS Initialization and Cleanup</title>
1135 A KMS device is abstracted and exposed as a set of planes, CRTCs, encoders
1136 and connectors. KMS drivers must thus create and initialize all those
1137 objects at load time after initializing mode setting.
1140 <title>CRTCs (struct <structname>drm_crtc</structname>)</title>
1142 A CRTC is an abstraction representing a part of the chip that contains a
1143 pointer to a scanout buffer. Therefore, the number of CRTCs available
1144 determines how many independent scanout buffers can be active at any
1145 given time. The CRTC structure contains several fields to support this:
1146 a pointer to some video memory (abstracted as a frame buffer object), a
1147 display mode, and an (x, y) offset into the video memory to support
1148 panning or configurations where one piece of video memory spans multiple
1152 <title>CRTC Initialization</title>
1154 A KMS device must create and register at least one struct
1155 <structname>drm_crtc</structname> instance. The instance is allocated
1156 and zeroed by the driver, possibly as part of a larger structure, and
1157 registered with a call to <function>drm_crtc_init</function> with a
1158 pointer to CRTC functions.
1162 <title>CRTC Operations</title>
1164 <title>Set Configuration</title>
1165 <synopsis>int (*set_config)(struct drm_mode_set *set);</synopsis>
1167 Apply a new CRTC configuration to the device. The configuration
1168 specifies a CRTC, a frame buffer to scan out from, a (x,y) position in
1169 the frame buffer, a display mode and an array of connectors to drive
1170 with the CRTC if possible.
1173 If the frame buffer specified in the configuration is NULL, the driver
1174 must detach all encoders connected to the CRTC and all connectors
1175 attached to those encoders and disable them.
1178 This operation is called with the mode config lock held.
1181 Note that the drm core has no notion of restoring the mode setting
1182 state after resume, since all resume handling is in the full
1183 responsibility of the driver. The common mode setting helper library
1184 though provides a helper which can be used for this:
1185 <function>drm_helper_resume_force_mode</function>.
1189 <title>Page Flipping</title>
1190 <synopsis>int (*page_flip)(struct drm_crtc *crtc, struct drm_framebuffer *fb,
1191 struct drm_pending_vblank_event *event);</synopsis>
1193 Schedule a page flip to the given frame buffer for the CRTC. This
1194 operation is called with the mode config mutex held.
1197 Page flipping is a synchronization mechanism that replaces the frame
1198 buffer being scanned out by the CRTC with a new frame buffer during
1199 vertical blanking, avoiding tearing. When an application requests a page
1200 flip the DRM core verifies that the new frame buffer is large enough to
1201 be scanned out by the CRTC in the currently configured mode and then
1202 calls the CRTC <methodname>page_flip</methodname> operation with a
1203 pointer to the new frame buffer.
1206 The <methodname>page_flip</methodname> operation schedules a page flip.
1207 Once any pending rendering targeting the new frame buffer has
1208 completed, the CRTC will be reprogrammed to display that frame buffer
1209 after the next vertical refresh. The operation must return immediately
1210 without waiting for rendering or page flip to complete and must block
1211 any new rendering to the frame buffer until the page flip completes.
1214 If a page flip can be successfully scheduled the driver must set the
1215 <code>drm_crtc-<fb</code> field to the new framebuffer pointed to
1216 by <code>fb</code>. This is important so that the reference counting
1217 on framebuffers stays balanced.
1220 If a page flip is already pending, the
1221 <methodname>page_flip</methodname> operation must return
1222 -<errorname>EBUSY</errorname>.
1225 To synchronize page flip to vertical blanking the driver will likely
1226 need to enable vertical blanking interrupts. It should call
1227 <function>drm_vblank_get</function> for that purpose, and call
1228 <function>drm_vblank_put</function> after the page flip completes.
1231 If the application has requested to be notified when page flip completes
1232 the <methodname>page_flip</methodname> operation will be called with a
1233 non-NULL <parameter>event</parameter> argument pointing to a
1234 <structname>drm_pending_vblank_event</structname> instance. Upon page
1235 flip completion the driver must call <methodname>drm_send_vblank_event</methodname>
1236 to fill in the event and send to wake up any waiting processes.
1237 This can be performed with
1238 <programlisting><![CDATA[
1239 spin_lock_irqsave(&dev->event_lock, flags);
1241 drm_send_vblank_event(dev, pipe, event);
1242 spin_unlock_irqrestore(&dev->event_lock, flags);
1243 ]]></programlisting>
1246 FIXME: Could drivers that don't need to wait for rendering to complete
1247 just add the event to <literal>dev->vblank_event_list</literal> and
1248 let the DRM core handle everything, as for "normal" vertical blanking
1252 While waiting for the page flip to complete, the
1253 <literal>event->base.link</literal> list head can be used freely by
1254 the driver to store the pending event in a driver-specific list.
1257 If the file handle is closed before the event is signaled, drivers must
1258 take care to destroy the event in their
1259 <methodname>preclose</methodname> operation (and, if needed, call
1260 <function>drm_vblank_put</function>).
1264 <title>Miscellaneous</title>
1267 <synopsis>void (*set_property)(struct drm_crtc *crtc,
1268 struct drm_property *property, uint64_t value);</synopsis>
1270 Set the value of the given CRTC property to
1271 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1272 for more information about properties.
1276 <synopsis>void (*gamma_set)(struct drm_crtc *crtc, u16 *r, u16 *g, u16 *b,
1277 uint32_t start, uint32_t size);</synopsis>
1279 Apply a gamma table to the device. The operation is optional.
1283 <synopsis>void (*destroy)(struct drm_crtc *crtc);</synopsis>
1285 Destroy the CRTC when not needed anymore. See
1286 <xref linkend="drm-kms-init"/>.
1294 <title>Planes (struct <structname>drm_plane</structname>)</title>
1296 A plane represents an image source that can be blended with or overlayed
1297 on top of a CRTC during the scanout process. Planes are associated with
1298 a frame buffer to crop a portion of the image memory (source) and
1299 optionally scale it to a destination size. The result is then blended
1300 with or overlayed on top of a CRTC.
1303 <title>Plane Initialization</title>
1305 Planes are optional. To create a plane, a KMS drivers allocates and
1306 zeroes an instances of struct <structname>drm_plane</structname>
1307 (possibly as part of a larger structure) and registers it with a call
1308 to <function>drm_plane_init</function>. The function takes a bitmask
1309 of the CRTCs that can be associated with the plane, a pointer to the
1310 plane functions and a list of format supported formats.
1314 <title>Plane Operations</title>
1317 <synopsis>int (*update_plane)(struct drm_plane *plane, struct drm_crtc *crtc,
1318 struct drm_framebuffer *fb, int crtc_x, int crtc_y,
1319 unsigned int crtc_w, unsigned int crtc_h,
1320 uint32_t src_x, uint32_t src_y,
1321 uint32_t src_w, uint32_t src_h);</synopsis>
1323 Enable and configure the plane to use the given CRTC and frame buffer.
1326 The source rectangle in frame buffer memory coordinates is given by
1327 the <parameter>src_x</parameter>, <parameter>src_y</parameter>,
1328 <parameter>src_w</parameter> and <parameter>src_h</parameter>
1329 parameters (as 16.16 fixed point values). Devices that don't support
1330 subpixel plane coordinates can ignore the fractional part.
1333 The destination rectangle in CRTC coordinates is given by the
1334 <parameter>crtc_x</parameter>, <parameter>crtc_y</parameter>,
1335 <parameter>crtc_w</parameter> and <parameter>crtc_h</parameter>
1336 parameters (as integer values). Devices scale the source rectangle to
1337 the destination rectangle. If scaling is not supported, and the source
1338 rectangle size doesn't match the destination rectangle size, the
1339 driver must return a -<errorname>EINVAL</errorname> error.
1343 <synopsis>int (*disable_plane)(struct drm_plane *plane);</synopsis>
1345 Disable the plane. The DRM core calls this method in response to a
1346 DRM_IOCTL_MODE_SETPLANE ioctl call with the frame buffer ID set to 0.
1347 Disabled planes must not be processed by the CRTC.
1351 <synopsis>void (*destroy)(struct drm_plane *plane);</synopsis>
1353 Destroy the plane when not needed anymore. See
1354 <xref linkend="drm-kms-init"/>.
1361 <title>Encoders (struct <structname>drm_encoder</structname>)</title>
1363 An encoder takes pixel data from a CRTC and converts it to a format
1364 suitable for any attached connectors. On some devices, it may be
1365 possible to have a CRTC send data to more than one encoder. In that
1366 case, both encoders would receive data from the same scanout buffer,
1367 resulting in a "cloned" display configuration across the connectors
1368 attached to each encoder.
1371 <title>Encoder Initialization</title>
1373 As for CRTCs, a KMS driver must create, initialize and register at
1374 least one struct <structname>drm_encoder</structname> instance. The
1375 instance is allocated and zeroed by the driver, possibly as part of a
1379 Drivers must initialize the struct <structname>drm_encoder</structname>
1380 <structfield>possible_crtcs</structfield> and
1381 <structfield>possible_clones</structfield> fields before registering the
1382 encoder. Both fields are bitmasks of respectively the CRTCs that the
1383 encoder can be connected to, and sibling encoders candidate for cloning.
1386 After being initialized, the encoder must be registered with a call to
1387 <function>drm_encoder_init</function>. The function takes a pointer to
1388 the encoder functions and an encoder type. Supported types are
1391 DRM_MODE_ENCODER_DAC for VGA and analog on DVI-I/DVI-A
1394 DRM_MODE_ENCODER_TMDS for DVI, HDMI and (embedded) DisplayPort
1397 DRM_MODE_ENCODER_LVDS for display panels
1400 DRM_MODE_ENCODER_TVDAC for TV output (Composite, S-Video, Component,
1404 DRM_MODE_ENCODER_VIRTUAL for virtual machine displays
1409 Encoders must be attached to a CRTC to be used. DRM drivers leave
1410 encoders unattached at initialization time. Applications (or the fbdev
1411 compatibility layer when implemented) are responsible for attaching the
1412 encoders they want to use to a CRTC.
1416 <title>Encoder Operations</title>
1419 <synopsis>void (*destroy)(struct drm_encoder *encoder);</synopsis>
1421 Called to destroy the encoder when not needed anymore. See
1422 <xref linkend="drm-kms-init"/>.
1426 <synopsis>void (*set_property)(struct drm_plane *plane,
1427 struct drm_property *property, uint64_t value);</synopsis>
1429 Set the value of the given plane property to
1430 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1431 for more information about properties.
1438 <title>Connectors (struct <structname>drm_connector</structname>)</title>
1440 A connector is the final destination for pixel data on a device, and
1441 usually connects directly to an external display device like a monitor
1442 or laptop panel. A connector can only be attached to one encoder at a
1443 time. The connector is also the structure where information about the
1444 attached display is kept, so it contains fields for display data, EDID
1445 data, DPMS & connection status, and information about modes
1446 supported on the attached displays.
1449 <title>Connector Initialization</title>
1451 Finally a KMS driver must create, initialize, register and attach at
1452 least one struct <structname>drm_connector</structname> instance. The
1453 instance is created as other KMS objects and initialized by setting the
1458 <term><structfield>interlace_allowed</structfield></term>
1460 Whether the connector can handle interlaced modes.
1464 <term><structfield>doublescan_allowed</structfield></term>
1466 Whether the connector can handle doublescan.
1470 <term><structfield>display_info
1471 </structfield></term>
1473 Display information is filled from EDID information when a display
1474 is detected. For non hot-pluggable displays such as flat panels in
1475 embedded systems, the driver should initialize the
1476 <structfield>display_info</structfield>.<structfield>width_mm</structfield>
1478 <structfield>display_info</structfield>.<structfield>height_mm</structfield>
1479 fields with the physical size of the display.
1483 <term id="drm-kms-connector-polled"><structfield>polled</structfield></term>
1485 Connector polling mode, a combination of
1488 <term>DRM_CONNECTOR_POLL_HPD</term>
1490 The connector generates hotplug events and doesn't need to be
1491 periodically polled. The CONNECT and DISCONNECT flags must not
1492 be set together with the HPD flag.
1496 <term>DRM_CONNECTOR_POLL_CONNECT</term>
1498 Periodically poll the connector for connection.
1502 <term>DRM_CONNECTOR_POLL_DISCONNECT</term>
1504 Periodically poll the connector for disconnection.
1508 Set to 0 for connectors that don't support connection status
1514 The connector is then registered with a call to
1515 <function>drm_connector_init</function> with a pointer to the connector
1516 functions and a connector type, and exposed through sysfs with a call to
1517 <function>drm_sysfs_connector_add</function>.
1520 Supported connector types are
1522 <listitem>DRM_MODE_CONNECTOR_VGA</listitem>
1523 <listitem>DRM_MODE_CONNECTOR_DVII</listitem>
1524 <listitem>DRM_MODE_CONNECTOR_DVID</listitem>
1525 <listitem>DRM_MODE_CONNECTOR_DVIA</listitem>
1526 <listitem>DRM_MODE_CONNECTOR_Composite</listitem>
1527 <listitem>DRM_MODE_CONNECTOR_SVIDEO</listitem>
1528 <listitem>DRM_MODE_CONNECTOR_LVDS</listitem>
1529 <listitem>DRM_MODE_CONNECTOR_Component</listitem>
1530 <listitem>DRM_MODE_CONNECTOR_9PinDIN</listitem>
1531 <listitem>DRM_MODE_CONNECTOR_DisplayPort</listitem>
1532 <listitem>DRM_MODE_CONNECTOR_HDMIA</listitem>
1533 <listitem>DRM_MODE_CONNECTOR_HDMIB</listitem>
1534 <listitem>DRM_MODE_CONNECTOR_TV</listitem>
1535 <listitem>DRM_MODE_CONNECTOR_eDP</listitem>
1536 <listitem>DRM_MODE_CONNECTOR_VIRTUAL</listitem>
1540 Connectors must be attached to an encoder to be used. For devices that
1541 map connectors to encoders 1:1, the connector should be attached at
1542 initialization time with a call to
1543 <function>drm_mode_connector_attach_encoder</function>. The driver must
1544 also set the <structname>drm_connector</structname>
1545 <structfield>encoder</structfield> field to point to the attached
1549 Finally, drivers must initialize the connectors state change detection
1550 with a call to <function>drm_kms_helper_poll_init</function>. If at
1551 least one connector is pollable but can't generate hotplug interrupts
1552 (indicated by the DRM_CONNECTOR_POLL_CONNECT and
1553 DRM_CONNECTOR_POLL_DISCONNECT connector flags), a delayed work will
1554 automatically be queued to periodically poll for changes. Connectors
1555 that can generate hotplug interrupts must be marked with the
1556 DRM_CONNECTOR_POLL_HPD flag instead, and their interrupt handler must
1557 call <function>drm_helper_hpd_irq_event</function>. The function will
1558 queue a delayed work to check the state of all connectors, but no
1559 periodic polling will be done.
1563 <title>Connector Operations</title>
1565 Unless otherwise state, all operations are mandatory.
1569 <synopsis>void (*dpms)(struct drm_connector *connector, int mode);</synopsis>
1571 The DPMS operation sets the power state of a connector. The mode
1574 <listitem><para>DRM_MODE_DPMS_ON</para></listitem>
1575 <listitem><para>DRM_MODE_DPMS_STANDBY</para></listitem>
1576 <listitem><para>DRM_MODE_DPMS_SUSPEND</para></listitem>
1577 <listitem><para>DRM_MODE_DPMS_OFF</para></listitem>
1581 In all but DPMS_ON mode the encoder to which the connector is attached
1582 should put the display in low-power mode by driving its signals
1583 appropriately. If more than one connector is attached to the encoder
1584 care should be taken not to change the power state of other displays as
1585 a side effect. Low-power mode should be propagated to the encoders and
1586 CRTCs when all related connectors are put in low-power mode.
1590 <title>Modes</title>
1591 <synopsis>int (*fill_modes)(struct drm_connector *connector, uint32_t max_width,
1592 uint32_t max_height);</synopsis>
1594 Fill the mode list with all supported modes for the connector. If the
1595 <parameter>max_width</parameter> and <parameter>max_height</parameter>
1596 arguments are non-zero, the implementation must ignore all modes wider
1597 than <parameter>max_width</parameter> or higher than
1598 <parameter>max_height</parameter>.
1601 The connector must also fill in this operation its
1602 <structfield>display_info</structfield>
1603 <structfield>width_mm</structfield> and
1604 <structfield>height_mm</structfield> fields with the connected display
1605 physical size in millimeters. The fields should be set to 0 if the value
1606 isn't known or is not applicable (for instance for projector devices).
1610 <title>Connection Status</title>
1612 The connection status is updated through polling or hotplug events when
1613 supported (see <xref linkend="drm-kms-connector-polled"/>). The status
1614 value is reported to userspace through ioctls and must not be used
1615 inside the driver, as it only gets initialized by a call to
1616 <function>drm_mode_getconnector</function> from userspace.
1618 <synopsis>enum drm_connector_status (*detect)(struct drm_connector *connector,
1619 bool force);</synopsis>
1621 Check to see if anything is attached to the connector. The
1622 <parameter>force</parameter> parameter is set to false whilst polling or
1623 to true when checking the connector due to user request.
1624 <parameter>force</parameter> can be used by the driver to avoid
1625 expensive, destructive operations during automated probing.
1628 Return connector_status_connected if something is connected to the
1629 connector, connector_status_disconnected if nothing is connected and
1630 connector_status_unknown if the connection state isn't known.
1633 Drivers should only return connector_status_connected if the connection
1634 status has really been probed as connected. Connectors that can't detect
1635 the connection status, or failed connection status probes, should return
1636 connector_status_unknown.
1640 <title>Miscellaneous</title>
1643 <synopsis>void (*set_property)(struct drm_connector *connector,
1644 struct drm_property *property, uint64_t value);</synopsis>
1646 Set the value of the given connector property to
1647 <parameter>value</parameter>. See <xref linkend="drm-kms-properties"/>
1648 for more information about properties.
1652 <synopsis>void (*destroy)(struct drm_connector *connector);</synopsis>
1654 Destroy the connector when not needed anymore. See
1655 <xref linkend="drm-kms-init"/>.
1663 <title>Cleanup</title>
1665 The DRM core manages its objects' lifetime. When an object is not needed
1666 anymore the core calls its destroy function, which must clean up and
1667 free every resource allocated for the object. Every
1668 <function>drm_*_init</function> call must be matched with a
1669 corresponding <function>drm_*_cleanup</function> call to cleanup CRTCs
1670 (<function>drm_crtc_cleanup</function>), planes
1671 (<function>drm_plane_cleanup</function>), encoders
1672 (<function>drm_encoder_cleanup</function>) and connectors
1673 (<function>drm_connector_cleanup</function>). Furthermore, connectors
1674 that have been added to sysfs must be removed by a call to
1675 <function>drm_sysfs_connector_remove</function> before calling
1676 <function>drm_connector_cleanup</function>.
1679 Connectors state change detection must be cleanup up with a call to
1680 <function>drm_kms_helper_poll_fini</function>.
1684 <title>Output discovery and initialization example</title>
1685 <programlisting><![CDATA[
1686 void intel_crt_init(struct drm_device *dev)
1688 struct drm_connector *connector;
1689 struct intel_output *intel_output;
1691 intel_output = kzalloc(sizeof(struct intel_output), GFP_KERNEL);
1695 connector = &intel_output->base;
1696 drm_connector_init(dev, &intel_output->base,
1697 &intel_crt_connector_funcs, DRM_MODE_CONNECTOR_VGA);
1699 drm_encoder_init(dev, &intel_output->enc, &intel_crt_enc_funcs,
1700 DRM_MODE_ENCODER_DAC);
1702 drm_mode_connector_attach_encoder(&intel_output->base,
1703 &intel_output->enc);
1705 /* Set up the DDC bus. */
1706 intel_output->ddc_bus = intel_i2c_create(dev, GPIOA, "CRTDDC_A");
1707 if (!intel_output->ddc_bus) {
1708 dev_printk(KERN_ERR, &dev->pdev->dev, "DDC bus registration "
1713 intel_output->type = INTEL_OUTPUT_ANALOG;
1714 connector->interlace_allowed = 0;
1715 connector->doublescan_allowed = 0;
1717 drm_encoder_helper_add(&intel_output->enc, &intel_crt_helper_funcs);
1718 drm_connector_helper_add(connector, &intel_crt_connector_helper_funcs);
1720 drm_sysfs_connector_add(connector);
1721 }]]></programlisting>
1723 In the example above (taken from the i915 driver), a CRTC, connector and
1724 encoder combination is created. A device-specific i2c bus is also
1725 created for fetching EDID data and performing monitor detection. Once
1726 the process is complete, the new connector is registered with sysfs to
1727 make its properties available to applications.
1731 <title>KMS API Functions</title>
1732 !Edrivers/gpu/drm/drm_crtc.c
1736 <!-- Internals: kms helper functions -->
1739 <title>Mode Setting Helper Functions</title>
1741 The CRTC, encoder and connector functions provided by the drivers
1742 implement the DRM API. They're called by the DRM core and ioctl handlers
1743 to handle device state changes and configuration request. As implementing
1744 those functions often requires logic not specific to drivers, mid-layer
1745 helper functions are available to avoid duplicating boilerplate code.
1748 The DRM core contains one mid-layer implementation. The mid-layer provides
1749 implementations of several CRTC, encoder and connector functions (called
1750 from the top of the mid-layer) that pre-process requests and call
1751 lower-level functions provided by the driver (at the bottom of the
1752 mid-layer). For instance, the
1753 <function>drm_crtc_helper_set_config</function> function can be used to
1754 fill the struct <structname>drm_crtc_funcs</structname>
1755 <structfield>set_config</structfield> field. When called, it will split
1756 the <methodname>set_config</methodname> operation in smaller, simpler
1757 operations and call the driver to handle them.
1760 To use the mid-layer, drivers call <function>drm_crtc_helper_add</function>,
1761 <function>drm_encoder_helper_add</function> and
1762 <function>drm_connector_helper_add</function> functions to install their
1763 mid-layer bottom operations handlers, and fill the
1764 <structname>drm_crtc_funcs</structname>,
1765 <structname>drm_encoder_funcs</structname> and
1766 <structname>drm_connector_funcs</structname> structures with pointers to
1767 the mid-layer top API functions. Installing the mid-layer bottom operation
1768 handlers is best done right after registering the corresponding KMS object.
1771 The mid-layer is not split between CRTC, encoder and connector operations.
1772 To use it, a driver must provide bottom functions for all of the three KMS
1776 <title>Helper Functions</title>
1779 <synopsis>int drm_crtc_helper_set_config(struct drm_mode_set *set);</synopsis>
1781 The <function>drm_crtc_helper_set_config</function> helper function
1782 is a CRTC <methodname>set_config</methodname> implementation. It
1783 first tries to locate the best encoder for each connector by calling
1784 the connector <methodname>best_encoder</methodname> helper
1788 After locating the appropriate encoders, the helper function will
1789 call the <methodname>mode_fixup</methodname> encoder and CRTC helper
1790 operations to adjust the requested mode, or reject it completely in
1791 which case an error will be returned to the application. If the new
1792 configuration after mode adjustment is identical to the current
1793 configuration the helper function will return without performing any
1797 If the adjusted mode is identical to the current mode but changes to
1798 the frame buffer need to be applied, the
1799 <function>drm_crtc_helper_set_config</function> function will call
1800 the CRTC <methodname>mode_set_base</methodname> helper operation. If
1801 the adjusted mode differs from the current mode, or if the
1802 <methodname>mode_set_base</methodname> helper operation is not
1803 provided, the helper function performs a full mode set sequence by
1804 calling the <methodname>prepare</methodname>,
1805 <methodname>mode_set</methodname> and
1806 <methodname>commit</methodname> CRTC and encoder helper operations,
1811 <synopsis>void drm_helper_connector_dpms(struct drm_connector *connector, int mode);</synopsis>
1813 The <function>drm_helper_connector_dpms</function> helper function
1814 is a connector <methodname>dpms</methodname> implementation that
1815 tracks power state of connectors. To use the function, drivers must
1816 provide <methodname>dpms</methodname> helper operations for CRTCs
1817 and encoders to apply the DPMS state to the device.
1820 The mid-layer doesn't track the power state of CRTCs and encoders.
1821 The <methodname>dpms</methodname> helper operations can thus be
1822 called with a mode identical to the currently active mode.
1826 <synopsis>int drm_helper_probe_single_connector_modes(struct drm_connector *connector,
1827 uint32_t maxX, uint32_t maxY);</synopsis>
1829 The <function>drm_helper_probe_single_connector_modes</function> helper
1830 function is a connector <methodname>fill_modes</methodname>
1831 implementation that updates the connection status for the connector
1832 and then retrieves a list of modes by calling the connector
1833 <methodname>get_modes</methodname> helper operation.
1836 The function filters out modes larger than
1837 <parameter>max_width</parameter> and <parameter>max_height</parameter>
1838 if specified. It then calls the connector
1839 <methodname>mode_valid</methodname> helper operation for each mode in
1840 the probed list to check whether the mode is valid for the connector.
1846 <title>CRTC Helper Operations</title>
1848 <listitem id="drm-helper-crtc-mode-fixup">
1849 <synopsis>bool (*mode_fixup)(struct drm_crtc *crtc,
1850 const struct drm_display_mode *mode,
1851 struct drm_display_mode *adjusted_mode);</synopsis>
1853 Let CRTCs adjust the requested mode or reject it completely. This
1854 operation returns true if the mode is accepted (possibly after being
1855 adjusted) or false if it is rejected.
1858 The <methodname>mode_fixup</methodname> operation should reject the
1859 mode if it can't reasonably use it. The definition of "reasonable"
1860 is currently fuzzy in this context. One possible behaviour would be
1861 to set the adjusted mode to the panel timings when a fixed-mode
1862 panel is used with hardware capable of scaling. Another behaviour
1863 would be to accept any input mode and adjust it to the closest mode
1864 supported by the hardware (FIXME: This needs to be clarified).
1868 <synopsis>int (*mode_set_base)(struct drm_crtc *crtc, int x, int y,
1869 struct drm_framebuffer *old_fb)</synopsis>
1871 Move the CRTC on the current frame buffer (stored in
1872 <literal>crtc->fb</literal>) to position (x,y). Any of the frame
1873 buffer, x position or y position may have been modified.
1876 This helper operation is optional. If not provided, the
1877 <function>drm_crtc_helper_set_config</function> function will fall
1878 back to the <methodname>mode_set</methodname> helper operation.
1881 FIXME: Why are x and y passed as arguments, as they can be accessed
1882 through <literal>crtc->x</literal> and
1883 <literal>crtc->y</literal>?
1887 <synopsis>void (*prepare)(struct drm_crtc *crtc);</synopsis>
1889 Prepare the CRTC for mode setting. This operation is called after
1890 validating the requested mode. Drivers use it to perform
1891 device-specific operations required before setting the new mode.
1895 <synopsis>int (*mode_set)(struct drm_crtc *crtc, struct drm_display_mode *mode,
1896 struct drm_display_mode *adjusted_mode, int x, int y,
1897 struct drm_framebuffer *old_fb);</synopsis>
1899 Set a new mode, position and frame buffer. Depending on the device
1900 requirements, the mode can be stored internally by the driver and
1901 applied in the <methodname>commit</methodname> operation, or
1902 programmed to the hardware immediately.
1905 The <methodname>mode_set</methodname> operation returns 0 on success
1906 or a negative error code if an error occurs.
1910 <synopsis>void (*commit)(struct drm_crtc *crtc);</synopsis>
1912 Commit a mode. This operation is called after setting the new mode.
1913 Upon return the device must use the new mode and be fully
1920 <title>Encoder Helper Operations</title>
1923 <synopsis>bool (*mode_fixup)(struct drm_encoder *encoder,
1924 const struct drm_display_mode *mode,
1925 struct drm_display_mode *adjusted_mode);</synopsis>
1927 Let encoders adjust the requested mode or reject it completely. This
1928 operation returns true if the mode is accepted (possibly after being
1929 adjusted) or false if it is rejected. See the
1930 <link linkend="drm-helper-crtc-mode-fixup">mode_fixup CRTC helper
1931 operation</link> for an explanation of the allowed adjustments.
1935 <synopsis>void (*prepare)(struct drm_encoder *encoder);</synopsis>
1937 Prepare the encoder for mode setting. This operation is called after
1938 validating the requested mode. Drivers use it to perform
1939 device-specific operations required before setting the new mode.
1943 <synopsis>void (*mode_set)(struct drm_encoder *encoder,
1944 struct drm_display_mode *mode,
1945 struct drm_display_mode *adjusted_mode);</synopsis>
1947 Set a new mode. Depending on the device requirements, the mode can
1948 be stored internally by the driver and applied in the
1949 <methodname>commit</methodname> operation, or programmed to the
1950 hardware immediately.
1954 <synopsis>void (*commit)(struct drm_encoder *encoder);</synopsis>
1956 Commit a mode. This operation is called after setting the new mode.
1957 Upon return the device must use the new mode and be fully
1964 <title>Connector Helper Operations</title>
1967 <synopsis>struct drm_encoder *(*best_encoder)(struct drm_connector *connector);</synopsis>
1969 Return a pointer to the best encoder for the connecter. Device that
1970 map connectors to encoders 1:1 simply return the pointer to the
1971 associated encoder. This operation is mandatory.
1975 <synopsis>int (*get_modes)(struct drm_connector *connector);</synopsis>
1977 Fill the connector's <structfield>probed_modes</structfield> list
1978 by parsing EDID data with <function>drm_add_edid_modes</function> or
1979 calling <function>drm_mode_probed_add</function> directly for every
1980 supported mode and return the number of modes it has detected. This
1981 operation is mandatory.
1984 When adding modes manually the driver creates each mode with a call to
1985 <function>drm_mode_create</function> and must fill the following fields.
1988 <synopsis>__u32 type;</synopsis>
1990 Mode type bitmask, a combination of
1993 <term>DRM_MODE_TYPE_BUILTIN</term>
1994 <listitem><para>not used?</para></listitem>
1997 <term>DRM_MODE_TYPE_CLOCK_C</term>
1998 <listitem><para>not used?</para></listitem>
2001 <term>DRM_MODE_TYPE_CRTC_C</term>
2002 <listitem><para>not used?</para></listitem>
2006 DRM_MODE_TYPE_PREFERRED - The preferred mode for the connector
2009 <para>not used?</para>
2013 <term>DRM_MODE_TYPE_DEFAULT</term>
2014 <listitem><para>not used?</para></listitem>
2017 <term>DRM_MODE_TYPE_USERDEF</term>
2018 <listitem><para>not used?</para></listitem>
2021 <term>DRM_MODE_TYPE_DRIVER</term>
2024 The mode has been created by the driver (as opposed to
2025 to user-created modes).
2030 Drivers must set the DRM_MODE_TYPE_DRIVER bit for all modes they
2031 create, and set the DRM_MODE_TYPE_PREFERRED bit for the preferred
2036 <synopsis>__u32 clock;</synopsis>
2037 <para>Pixel clock frequency in kHz unit</para>
2040 <synopsis>__u16 hdisplay, hsync_start, hsync_end, htotal;
2041 __u16 vdisplay, vsync_start, vsync_end, vtotal;</synopsis>
2042 <para>Horizontal and vertical timing information</para>
2044 Active Front Sync Back
2046 <-----------------------><----------------><-------------><-------------->
2048 //////////////////////|
2049 ////////////////////// |
2050 ////////////////////// |.................. ................
2053 <----- [hv]display ----->
2054 <------------- [hv]sync_start ------------>
2055 <--------------------- [hv]sync_end --------------------->
2056 <-------------------------------- [hv]total ----------------------------->
2060 <synopsis>__u16 hskew;
2061 __u16 vscan;</synopsis>
2062 <para>Unknown</para>
2065 <synopsis>__u32 flags;</synopsis>
2067 Mode flags, a combination of
2070 <term>DRM_MODE_FLAG_PHSYNC</term>
2072 Horizontal sync is active high
2076 <term>DRM_MODE_FLAG_NHSYNC</term>
2078 Horizontal sync is active low
2082 <term>DRM_MODE_FLAG_PVSYNC</term>
2084 Vertical sync is active high
2088 <term>DRM_MODE_FLAG_NVSYNC</term>
2090 Vertical sync is active low
2094 <term>DRM_MODE_FLAG_INTERLACE</term>
2100 <term>DRM_MODE_FLAG_DBLSCAN</term>
2102 Mode uses doublescan
2106 <term>DRM_MODE_FLAG_CSYNC</term>
2108 Mode uses composite sync
2112 <term>DRM_MODE_FLAG_PCSYNC</term>
2114 Composite sync is active high
2118 <term>DRM_MODE_FLAG_NCSYNC</term>
2120 Composite sync is active low
2124 <term>DRM_MODE_FLAG_HSKEW</term>
2126 hskew provided (not used?)
2130 <term>DRM_MODE_FLAG_BCAST</term>
2136 <term>DRM_MODE_FLAG_PIXMUX</term>
2142 <term>DRM_MODE_FLAG_DBLCLK</term>
2148 <term>DRM_MODE_FLAG_CLKDIV2</term>
2156 Note that modes marked with the INTERLACE or DBLSCAN flags will be
2158 <function>drm_helper_probe_single_connector_modes</function> if
2159 the connector's <structfield>interlace_allowed</structfield> or
2160 <structfield>doublescan_allowed</structfield> field is set to 0.
2164 <synopsis>char name[DRM_DISPLAY_MODE_LEN];</synopsis>
2166 Mode name. The driver must call
2167 <function>drm_mode_set_name</function> to fill the mode name from
2168 <structfield>hdisplay</structfield>,
2169 <structfield>vdisplay</structfield> and interlace flag after
2170 filling the corresponding fields.
2176 The <structfield>vrefresh</structfield> value is computed by
2177 <function>drm_helper_probe_single_connector_modes</function>.
2180 When parsing EDID data, <function>drm_add_edid_modes</function> fill the
2181 connector <structfield>display_info</structfield>
2182 <structfield>width_mm</structfield> and
2183 <structfield>height_mm</structfield> fields. When creating modes
2184 manually the <methodname>get_modes</methodname> helper operation must
2185 set the <structfield>display_info</structfield>
2186 <structfield>width_mm</structfield> and
2187 <structfield>height_mm</structfield> fields if they haven't been set
2188 already (for instance at initialization time when a fixed-size panel is
2189 attached to the connector). The mode <structfield>width_mm</structfield>
2190 and <structfield>height_mm</structfield> fields are only used internally
2191 during EDID parsing and should not be set when creating modes manually.
2195 <synopsis>int (*mode_valid)(struct drm_connector *connector,
2196 struct drm_display_mode *mode);</synopsis>
2198 Verify whether a mode is valid for the connector. Return MODE_OK for
2199 supported modes and one of the enum drm_mode_status values (MODE_*)
2200 for unsupported modes. This operation is mandatory.
2203 As the mode rejection reason is currently not used beside for
2204 immediately removing the unsupported mode, an implementation can
2205 return MODE_BAD regardless of the exact reason why the mode is not
2209 Note that the <methodname>mode_valid</methodname> helper operation is
2210 only called for modes detected by the device, and
2211 <emphasis>not</emphasis> for modes set by the user through the CRTC
2212 <methodname>set_config</methodname> operation.
2218 <title>Modeset Helper Functions Reference</title>
2219 !Edrivers/gpu/drm/drm_crtc_helper.c
2222 <title>fbdev Helper Functions Reference</title>
2223 !Pdrivers/gpu/drm/drm_fb_helper.c fbdev helpers
2224 !Edrivers/gpu/drm/drm_fb_helper.c
2225 !Iinclude/drm/drm_fb_helper.h
2228 <title>Display Port Helper Functions Reference</title>
2229 !Pdrivers/gpu/drm/drm_dp_helper.c dp helpers
2230 !Iinclude/drm/drm_dp_helper.h
2231 !Edrivers/gpu/drm/drm_dp_helper.c
2234 <title>EDID Helper Functions Reference</title>
2235 !Edrivers/gpu/drm/drm_edid.c
2238 <title>Rectangle Utilities Reference</title>
2239 !Pinclude/drm/drm_rect.h rect utils
2240 !Iinclude/drm/drm_rect.h
2241 !Edrivers/gpu/drm/drm_rect.c
2244 <title>Flip-work Helper Reference</title>
2245 !Pinclude/drm/drm_flip_work.h flip utils
2246 !Iinclude/drm/drm_flip_work.h
2247 !Edrivers/gpu/drm/drm_flip_work.c
2250 <title>HDMI Infoframes Helper Reference</title>
2252 Strictly speaking this is not a DRM helper library but generally useable
2253 by any driver interfacing with HDMI outputs like v4l or alsa drivers.
2254 But it nicely fits into the overall topic of mode setting helper
2255 libraries and hence is also included here.
2257 !Iinclude/linux/hdmi.h
2258 !Edrivers/video/hdmi.c
2262 <!-- Internals: kms properties -->
2264 <sect1 id="drm-kms-properties">
2265 <title>KMS Properties</title>
2267 Drivers may need to expose additional parameters to applications than
2268 those described in the previous sections. KMS supports attaching
2269 properties to CRTCs, connectors and planes and offers a userspace API to
2270 list, get and set the property values.
2273 Properties are identified by a name that uniquely defines the property
2274 purpose, and store an associated value. For all property types except blob
2275 properties the value is a 64-bit unsigned integer.
2278 KMS differentiates between properties and property instances. Drivers
2279 first create properties and then create and associate individual instances
2280 of those properties to objects. A property can be instantiated multiple
2281 times and associated with different objects. Values are stored in property
2282 instances, and all other property information are stored in the propery
2283 and shared between all instances of the property.
2286 Every property is created with a type that influences how the KMS core
2287 handles the property. Supported property types are
2290 <term>DRM_MODE_PROP_RANGE</term>
2291 <listitem><para>Range properties report their minimum and maximum
2292 admissible values. The KMS core verifies that values set by
2293 application fit in that range.</para></listitem>
2296 <term>DRM_MODE_PROP_ENUM</term>
2297 <listitem><para>Enumerated properties take a numerical value that
2298 ranges from 0 to the number of enumerated values defined by the
2299 property minus one, and associate a free-formed string name to each
2300 value. Applications can retrieve the list of defined value-name pairs
2301 and use the numerical value to get and set property instance values.
2305 <term>DRM_MODE_PROP_BITMASK</term>
2306 <listitem><para>Bitmask properties are enumeration properties that
2307 additionally restrict all enumerated values to the 0..63 range.
2308 Bitmask property instance values combine one or more of the
2309 enumerated bits defined by the property.</para></listitem>
2312 <term>DRM_MODE_PROP_BLOB</term>
2313 <listitem><para>Blob properties store a binary blob without any format
2314 restriction. The binary blobs are created as KMS standalone objects,
2315 and blob property instance values store the ID of their associated
2317 <para>Blob properties are only used for the connector EDID property
2318 and cannot be created by drivers.</para></listitem>
2323 To create a property drivers call one of the following functions depending
2324 on the property type. All property creation functions take property flags
2325 and name, as well as type-specific arguments.
2328 <synopsis>struct drm_property *drm_property_create_range(struct drm_device *dev, int flags,
2330 uint64_t min, uint64_t max);</synopsis>
2331 <para>Create a range property with the given minimum and maximum
2335 <synopsis>struct drm_property *drm_property_create_enum(struct drm_device *dev, int flags,
2337 const struct drm_prop_enum_list *props,
2338 int num_values);</synopsis>
2339 <para>Create an enumerated property. The <parameter>props</parameter>
2340 argument points to an array of <parameter>num_values</parameter>
2341 value-name pairs.</para>
2344 <synopsis>struct drm_property *drm_property_create_bitmask(struct drm_device *dev,
2345 int flags, const char *name,
2346 const struct drm_prop_enum_list *props,
2347 int num_values);</synopsis>
2348 <para>Create a bitmask property. The <parameter>props</parameter>
2349 argument points to an array of <parameter>num_values</parameter>
2350 value-name pairs.</para>
2355 Properties can additionally be created as immutable, in which case they
2356 will be read-only for applications but can be modified by the driver. To
2357 create an immutable property drivers must set the DRM_MODE_PROP_IMMUTABLE
2358 flag at property creation time.
2361 When no array of value-name pairs is readily available at property
2362 creation time for enumerated or range properties, drivers can create
2363 the property using the <function>drm_property_create</function> function
2364 and manually add enumeration value-name pairs by calling the
2365 <function>drm_property_add_enum</function> function. Care must be taken to
2366 properly specify the property type through the <parameter>flags</parameter>
2370 After creating properties drivers can attach property instances to CRTC,
2371 connector and plane objects by calling the
2372 <function>drm_object_attach_property</function>. The function takes a
2373 pointer to the target object, a pointer to the previously created property
2374 and an initial instance value.
2378 <!-- Internals: vertical blanking -->
2380 <sect1 id="drm-vertical-blank">
2381 <title>Vertical Blanking</title>
2383 Vertical blanking plays a major role in graphics rendering. To achieve
2384 tear-free display, users must synchronize page flips and/or rendering to
2385 vertical blanking. The DRM API offers ioctls to perform page flips
2386 synchronized to vertical blanking and wait for vertical blanking.
2389 The DRM core handles most of the vertical blanking management logic, which
2390 involves filtering out spurious interrupts, keeping race-free blanking
2391 counters, coping with counter wrap-around and resets and keeping use
2392 counts. It relies on the driver to generate vertical blanking interrupts
2393 and optionally provide a hardware vertical blanking counter. Drivers must
2394 implement the following operations.
2398 <synopsis>int (*enable_vblank) (struct drm_device *dev, int crtc);
2399 void (*disable_vblank) (struct drm_device *dev, int crtc);</synopsis>
2401 Enable or disable vertical blanking interrupts for the given CRTC.
2405 <synopsis>u32 (*get_vblank_counter) (struct drm_device *dev, int crtc);</synopsis>
2407 Retrieve the value of the vertical blanking counter for the given
2408 CRTC. If the hardware maintains a vertical blanking counter its value
2409 should be returned. Otherwise drivers can use the
2410 <function>drm_vblank_count</function> helper function to handle this
2416 Drivers must initialize the vertical blanking handling core with a call to
2417 <function>drm_vblank_init</function> in their
2418 <methodname>load</methodname> operation. The function will set the struct
2419 <structname>drm_device</structname>
2420 <structfield>vblank_disable_allowed</structfield> field to 0. This will
2421 keep vertical blanking interrupts enabled permanently until the first mode
2422 set operation, where <structfield>vblank_disable_allowed</structfield> is
2423 set to 1. The reason behind this is not clear. Drivers can set the field
2424 to 1 after <function>calling drm_vblank_init</function> to make vertical
2425 blanking interrupts dynamically managed from the beginning.
2428 Vertical blanking interrupts can be enabled by the DRM core or by drivers
2429 themselves (for instance to handle page flipping operations). The DRM core
2430 maintains a vertical blanking use count to ensure that the interrupts are
2431 not disabled while a user still needs them. To increment the use count,
2432 drivers call <function>drm_vblank_get</function>. Upon return vertical
2433 blanking interrupts are guaranteed to be enabled.
2436 To decrement the use count drivers call
2437 <function>drm_vblank_put</function>. Only when the use count drops to zero
2438 will the DRM core disable the vertical blanking interrupts after a delay
2439 by scheduling a timer. The delay is accessible through the vblankoffdelay
2440 module parameter or the <varname>drm_vblank_offdelay</varname> global
2441 variable and expressed in milliseconds. Its default value is 5000 ms.
2444 When a vertical blanking interrupt occurs drivers only need to call the
2445 <function>drm_handle_vblank</function> function to account for the
2449 Resources allocated by <function>drm_vblank_init</function> must be freed
2450 with a call to <function>drm_vblank_cleanup</function> in the driver
2451 <methodname>unload</methodname> operation handler.
2455 <!-- Internals: open/close, file operations and ioctls -->
2458 <title>Open/Close, File Operations and IOCTLs</title>
2460 <title>Open and Close</title>
2461 <synopsis>int (*firstopen) (struct drm_device *);
2462 void (*lastclose) (struct drm_device *);
2463 int (*open) (struct drm_device *, struct drm_file *);
2464 void (*preclose) (struct drm_device *, struct drm_file *);
2465 void (*postclose) (struct drm_device *, struct drm_file *);</synopsis>
2466 <abstract>Open and close handlers. None of those methods are mandatory.
2469 The <methodname>firstopen</methodname> method is called by the DRM core
2470 for legacy UMS (User Mode Setting) drivers only when an application
2471 opens a device that has no other opened file handle. UMS drivers can
2472 implement it to acquire device resources. KMS drivers can't use the
2473 method and must acquire resources in the <methodname>load</methodname>
2477 Similarly the <methodname>lastclose</methodname> method is called when
2478 the last application holding a file handle opened on the device closes
2479 it, for both UMS and KMS drivers. Additionally, the method is also
2480 called at module unload time or, for hot-pluggable devices, when the
2481 device is unplugged. The <methodname>firstopen</methodname> and
2482 <methodname>lastclose</methodname> calls can thus be unbalanced.
2485 The <methodname>open</methodname> method is called every time the device
2486 is opened by an application. Drivers can allocate per-file private data
2487 in this method and store them in the struct
2488 <structname>drm_file</structname> <structfield>driver_priv</structfield>
2489 field. Note that the <methodname>open</methodname> method is called
2490 before <methodname>firstopen</methodname>.
2493 The close operation is split into <methodname>preclose</methodname> and
2494 <methodname>postclose</methodname> methods. Drivers must stop and
2495 cleanup all per-file operations in the <methodname>preclose</methodname>
2496 method. For instance pending vertical blanking and page flip events must
2497 be cancelled. No per-file operation is allowed on the file handle after
2498 returning from the <methodname>preclose</methodname> method.
2501 Finally the <methodname>postclose</methodname> method is called as the
2502 last step of the close operation, right before calling the
2503 <methodname>lastclose</methodname> method if no other open file handle
2504 exists for the device. Drivers that have allocated per-file private data
2505 in the <methodname>open</methodname> method should free it here.
2508 The <methodname>lastclose</methodname> method should restore CRTC and
2509 plane properties to default value, so that a subsequent open of the
2510 device will not inherit state from the previous user. It can also be
2511 used to execute delayed power switching state changes, e.g. in
2512 conjunction with the vga-switcheroo infrastructure. Beyond that KMS
2513 drivers should not do any further cleanup. Only legacy UMS drivers might
2514 need to clean up device state so that the vga console or an independent
2515 fbdev driver could take over.
2519 <title>File Operations</title>
2520 <synopsis>const struct file_operations *fops</synopsis>
2521 <abstract>File operations for the DRM device node.</abstract>
2523 Drivers must define the file operations structure that forms the DRM
2524 userspace API entry point, even though most of those operations are
2525 implemented in the DRM core. The <methodname>open</methodname>,
2526 <methodname>release</methodname> and <methodname>ioctl</methodname>
2527 operations are handled by
2529 .owner = THIS_MODULE,
2531 .release = drm_release,
2532 .unlocked_ioctl = drm_ioctl,
2533 #ifdef CONFIG_COMPAT
2534 .compat_ioctl = drm_compat_ioctl,
2539 Drivers that implement private ioctls that requires 32/64bit
2540 compatibility support must provide their own
2541 <methodname>compat_ioctl</methodname> handler that processes private
2542 ioctls and calls <function>drm_compat_ioctl</function> for core ioctls.
2545 The <methodname>read</methodname> and <methodname>poll</methodname>
2546 operations provide support for reading DRM events and polling them. They
2551 .llseek = no_llseek,
2555 The memory mapping implementation varies depending on how the driver
2556 manages memory. Pre-GEM drivers will use <function>drm_mmap</function>,
2557 while GEM-aware drivers will use <function>drm_gem_mmap</function>. See
2558 <xref linkend="drm-gem"/>.
2560 .mmap = drm_gem_mmap,
2564 No other file operation is supported by the DRM API.
2568 <title>IOCTLs</title>
2569 <synopsis>struct drm_ioctl_desc *ioctls;
2570 int num_ioctls;</synopsis>
2571 <abstract>Driver-specific ioctls descriptors table.</abstract>
2573 Driver-specific ioctls numbers start at DRM_COMMAND_BASE. The ioctls
2574 descriptors table is indexed by the ioctl number offset from the base
2575 value. Drivers can use the DRM_IOCTL_DEF_DRV() macro to initialize the
2579 <programlisting>DRM_IOCTL_DEF_DRV(ioctl, func, flags)</programlisting>
2581 <parameter>ioctl</parameter> is the ioctl name. Drivers must define
2582 the DRM_##ioctl and DRM_IOCTL_##ioctl macros to the ioctl number
2583 offset from DRM_COMMAND_BASE and the ioctl number respectively. The
2584 first macro is private to the device while the second must be exposed
2585 to userspace in a public header.
2588 <parameter>func</parameter> is a pointer to the ioctl handler function
2589 compatible with the <type>drm_ioctl_t</type> type.
2590 <programlisting>typedef int drm_ioctl_t(struct drm_device *dev, void *data,
2591 struct drm_file *file_priv);</programlisting>
2594 <parameter>flags</parameter> is a bitmask combination of the following
2595 values. It restricts how the ioctl is allowed to be called.
2598 DRM_AUTH - Only authenticated callers allowed
2601 DRM_MASTER - The ioctl can only be called on the master file
2605 DRM_ROOT_ONLY - Only callers with the SYSADMIN capability allowed
2608 DRM_CONTROL_ALLOW - The ioctl can only be called on a control
2612 DRM_UNLOCKED - The ioctl handler will be called without locking
2613 the DRM global mutex
2621 <title>Legacy Support Code</title>
2623 The section very brievely covers some of the old legacy support code which
2624 is only used by old DRM drivers which have done a so-called shadow-attach
2625 to the underlying device instead of registering as a real driver. This
2626 also includes some of the old generic buffer mangement and command
2627 submission code. Do not use any of this in new and modern drivers.
2631 <title>Legacy Suspend/Resume</title>
2633 The DRM core provides some suspend/resume code, but drivers wanting full
2634 suspend/resume support should provide save() and restore() functions.
2635 These are called at suspend, hibernate, or resume time, and should perform
2636 any state save or restore required by your device across suspend or
2639 <synopsis>int (*suspend) (struct drm_device *, pm_message_t state);
2640 int (*resume) (struct drm_device *);</synopsis>
2642 Those are legacy suspend and resume methods which
2643 <emphasis>only</emphasis> work with the legacy shadow-attach driver
2644 registration functions. New driver should use the power management
2645 interface provided by their bus type (usually through
2646 the struct <structname>device_driver</structname> dev_pm_ops) and set
2647 these methods to NULL.
2652 <title>Legacy DMA Services</title>
2654 This should cover how DMA mapping etc. is supported by the core.
2655 These functions are deprecated and should not be used.
2664 - Document the struct_mutex catch-all lock
2665 - Document connector properties
2667 - Why is the load method optional?
2668 - What are drivers supposed to set the initial display state to, and how?
2669 Connector's DPMS states are not initialized and are thus equal to
2670 DRM_MODE_DPMS_ON. The fbcon compatibility layer calls
2671 drm_helper_disable_unused_functions(), which disables unused encoders and
2672 CRTCs, but doesn't touch the connectors' DPMS state, and
2673 drm_helper_connector_dpms() in reaction to fbdev blanking events. Do drivers
2674 that don't implement (or just don't use) fbcon compatibility need to call
2675 those functions themselves?
2676 - KMS drivers must call drm_vblank_pre_modeset() and drm_vblank_post_modeset()
2677 around mode setting. Should this be done in the DRM core?
2678 - vblank_disable_allowed is set to 1 in the first drm_vblank_post_modeset()
2679 call and never set back to 0. It seems to be safe to permanently set it to 1
2680 in drm_vblank_init() for KMS driver, and it might be safe for UMS drivers as
2681 well. This should be investigated.
2682 - crtc and connector .save and .restore operations are only used internally in
2683 drivers, should they be removed from the core?
2684 - encoder mid-layer .save and .restore operations are only used internally in
2685 drivers, should they be removed from the core?
2686 - encoder mid-layer .detect operation is only used internally in drivers,
2687 should it be removed from the core?
2690 <!-- External interfaces -->
2692 <chapter id="drmExternals">
2693 <title>Userland interfaces</title>
2695 The DRM core exports several interfaces to applications,
2696 generally intended to be used through corresponding libdrm
2697 wrapper functions. In addition, drivers export device-specific
2698 interfaces for use by userspace drivers & device-aware
2699 applications through ioctls and sysfs files.
2702 External interfaces include: memory mapping, context management,
2703 DMA operations, AGP management, vblank control, fence
2704 management, memory management, and output management.
2707 Cover generic ioctls and sysfs layout here. We only need high-level
2708 info, since man pages should cover the rest.
2711 <!-- External: render nodes -->
2714 <title>Render nodes</title>
2716 DRM core provides multiple character-devices for user-space to use.
2717 Depending on which device is opened, user-space can perform a different
2718 set of operations (mainly ioctls). The primary node is always created
2719 and called card<num>. Additionally, a currently
2720 unused control node, called controlD<num> is also
2721 created. The primary node provides all legacy operations and
2722 historically was the only interface used by userspace. With KMS, the
2723 control node was introduced. However, the planned KMS control interface
2724 has never been written and so the control node stays unused to date.
2727 With the increased use of offscreen renderers and GPGPU applications,
2728 clients no longer require running compositors or graphics servers to
2729 make use of a GPU. But the DRM API required unprivileged clients to
2730 authenticate to a DRM-Master prior to getting GPU access. To avoid this
2731 step and to grant clients GPU access without authenticating, render
2732 nodes were introduced. Render nodes solely serve render clients, that
2733 is, no modesetting or privileged ioctls can be issued on render nodes.
2734 Only non-global rendering commands are allowed. If a driver supports
2735 render nodes, it must advertise it via the DRIVER_RENDER
2736 DRM driver capability. If not supported, the primary node must be used
2737 for render clients together with the legacy drmAuth authentication
2741 If a driver advertises render node support, DRM core will create a
2742 separate render node called renderD<num>. There will
2743 be one render node per device. No ioctls except PRIME-related ioctls
2744 will be allowed on this node. Especially GEM_OPEN will be
2745 explicitly prohibited. Render nodes are designed to avoid the
2746 buffer-leaks, which occur if clients guess the flink names or mmap
2747 offsets on the legacy interface. Additionally to this basic interface,
2748 drivers must mark their driver-dependent render-only ioctls as
2749 DRM_RENDER_ALLOW so render clients can use them. Driver
2750 authors must be careful not to allow any privileged ioctls on render
2754 With render nodes, user-space can now control access to the render node
2755 via basic file-system access-modes. A running graphics server which
2756 authenticates clients on the privileged primary/legacy node is no longer
2757 required. Instead, a client can open the render node and is immediately
2758 granted GPU access. Communication between clients (or servers) is done
2759 via PRIME. FLINK from render node to legacy node is not supported. New
2760 clients must not use the insecure FLINK interface.
2763 Besides dropping all modeset/global ioctls, render nodes also drop the
2764 DRM-Master concept. There is no reason to associate render clients with
2765 a DRM-Master as they are independent of any graphics server. Besides,
2766 they must work without any running master, anyway.
2767 Drivers must be able to run without a master object if they support
2768 render nodes. If, on the other hand, a driver requires shared state
2769 between clients which is visible to user-space and accessible beyond
2770 open-file boundaries, they cannot support render nodes.
2774 <!-- External: vblank handling -->
2777 <title>VBlank event handling</title>
2779 The DRM core exposes two vertical blank related ioctls:
2782 <term>DRM_IOCTL_WAIT_VBLANK</term>
2785 This takes a struct drm_wait_vblank structure as its argument,
2786 and it is used to block or request a signal when a specified
2787 vblank event occurs.
2792 <term>DRM_IOCTL_MODESET_CTL</term>
2795 This should be called by application level drivers before and
2796 after mode setting, since on many devices the vertical blank
2797 counter is reset at that time. Internally, the DRM snapshots
2798 the last vblank count when the ioctl is called with the
2799 _DRM_PRE_MODESET command, so that the counter won't go backwards
2800 (which is dealt with when _DRM_POST_MODESET is used).
2805 <!--!Edrivers/char/drm/drm_irq.c-->
2811 <part id="drmDrivers">
2812 <title>DRM Drivers</title>
2816 This second part of the DRM Developer's Guide documents driver code,
2817 implementation details and also all the driver-specific userspace
2818 interfaces. Especially since all hardware-acceleration interfaces to
2819 userspace are driver specific for efficiency and other reasons these
2820 interfaces can be rather substantial. Hence every driver has its own
2825 <chapter id="drmI915">
2826 <title>drm/i915 Intel GFX Driver</title>
2828 The drm/i915 driver supports all (with the exception of some very early
2829 models) integrated GFX chipsets with both Intel display and rendering
2830 blocks. This excludes a set of SoC platforms with an SGX rendering unit,
2831 those have basic support through the gma500 drm driver.
2834 <title>Display Hardware Handling</title>
2836 This section covers everything related to the display hardware including
2837 the mode setting infrastructure, plane, sprite and cursor handling and
2838 display, output probing and related topics.
2841 <title>Mode Setting Infrastructure</title>
2843 The i915 driver is thus far the only DRM driver which doesn't use the
2844 common DRM helper code to implement mode setting sequences. Thus it
2845 has its own tailor-made infrastructure for executing a display
2846 configuration change.
2850 <title>Plane Configuration</title>
2852 This section covers plane configuration and composition with the
2853 primary plane, sprites, cursors and overlays. This includes the
2854 infrastructure to do atomic vsync'ed updates of all this state and
2855 also tightly coupled topics like watermark setup and computation,
2856 framebuffer compression and panel self refresh.
2860 <title>Output Probing</title>
2862 This section covers output probing and related infrastructure like the
2863 hotplug interrupt storm detection and mitigation code. Note that the
2864 i915 driver still uses most of the common DRM helper code for output
2865 probing, so those sections fully apply.
2871 <title>Memory Management and Command Submission</title>
2873 This sections covers all things related to the GEM implementation in the
projects / firefly-linux-kernel-4.4.55.git / blob