xref: /linux/Documentation/gpu/drm-mm.rst (revision ca220141fa8ebae09765a242076b2b77338106b0)
1=====================
2DRM Memory Management
3=====================
4
5Modern Linux systems require large amount of graphics memory to store
6frame buffers, textures, vertices and other graphics-related data. Given
7the very dynamic nature of many of that data, managing graphics memory
8efficiently is thus crucial for the graphics stack and plays a central
9role in the DRM infrastructure.
10
11The DRM core includes two memory managers, namely Translation Table Manager
12(TTM) and Graphics Execution Manager (GEM). TTM was the first DRM memory
13manager to be developed and tried to be a one-size-fits-them all
14solution. It provides a single userspace API to accommodate the need of
15all hardware, supporting both Unified Memory Architecture (UMA) devices
16and devices with dedicated video RAM (i.e. most discrete video cards).
17This resulted in a large, complex piece of code that turned out to be
18hard to use for driver development.
19
20GEM started as an Intel-sponsored project in reaction to TTM's
21complexity. Its design philosophy is completely different: instead of
22providing a solution to every graphics memory-related problems, GEM
23identified common code between drivers and created a support library to
24share it. GEM has simpler initialization and execution requirements than
25TTM, but has no video RAM management capabilities and is thus limited to
26UMA devices.
27
28The Translation Table Manager (TTM)
29===================================
30
31.. kernel-doc:: drivers/gpu/drm/ttm/ttm_module.c
32   :doc: TTM
33
34.. kernel-doc:: include/drm/ttm/ttm_caching.h
35   :internal:
36
37TTM device object reference
38---------------------------
39
40.. kernel-doc:: include/drm/ttm/ttm_device.h
41   :internal:
42
43.. kernel-doc:: drivers/gpu/drm/ttm/ttm_device.c
44   :export:
45
46TTM resource placement reference
47--------------------------------
48
49.. kernel-doc:: include/drm/ttm/ttm_placement.h
50   :internal:
51
52TTM resource object reference
53-----------------------------
54
55.. kernel-doc:: include/drm/ttm/ttm_resource.h
56   :internal:
57
58.. kernel-doc:: drivers/gpu/drm/ttm/ttm_resource.c
59   :export:
60
61TTM TT object reference
62-----------------------
63
64.. kernel-doc:: include/drm/ttm/ttm_tt.h
65   :internal:
66
67.. kernel-doc:: drivers/gpu/drm/ttm/ttm_tt.c
68   :export:
69
70TTM page pool reference
71-----------------------
72
73.. kernel-doc:: include/drm/ttm/ttm_pool.h
74   :internal:
75
76.. kernel-doc:: drivers/gpu/drm/ttm/ttm_pool.c
77   :export:
78
79The Graphics Execution Manager (GEM)
80====================================
81
82The GEM design approach has resulted in a memory manager that doesn't
83provide full coverage of all (or even all common) use cases in its
84userspace or kernel API. GEM exposes a set of standard memory-related
85operations to userspace and a set of helper functions to drivers, and
86let drivers implement hardware-specific operations with their own
87private API.
88
89The GEM userspace API is described in the `GEM - the Graphics Execution
90Manager <http://lwn.net/Articles/283798/>`__ article on LWN. While
91slightly outdated, the document provides a good overview of the GEM API
92principles. Buffer allocation and read and write operations, described
93as part of the common GEM API, are currently implemented using
94driver-specific ioctls.
95
96GEM is data-agnostic. It manages abstract buffer objects without knowing
97what individual buffers contain. APIs that require knowledge of buffer
98contents or purpose, such as buffer allocation or synchronization
99primitives, are thus outside of the scope of GEM and must be implemented
100using driver-specific ioctls.
101
102On a fundamental level, GEM involves several operations:
103
104-  Memory allocation and freeing
105-  Command execution
106-  Aperture management at command execution time
107
108Buffer object allocation is relatively straightforward and largely
109provided by Linux's shmem layer, which provides memory to back each
110object.
111
112Device-specific operations, such as command execution, pinning, buffer
113read & write, mapping, and domain ownership transfers are left to
114driver-specific ioctls.
115
116GEM Initialization
117------------------
118
119Drivers that use GEM must set the DRIVER_GEM bit in the struct
120:c:type:`struct drm_driver <drm_driver>` driver_features
121field. The DRM core will then automatically initialize the GEM core
122before calling the load operation. Behind the scene, this will create a
123DRM Memory Manager object which provides an address space pool for
124object allocation.
125
126In a KMS configuration, drivers need to allocate and initialize a
127command ring buffer following core GEM initialization if required by the
128hardware. UMA devices usually have what is called a "stolen" memory
129region, which provides space for the initial framebuffer and large,
130contiguous memory regions required by the device. This space is
131typically not managed by GEM, and must be initialized separately into
132its own DRM MM object.
133
134GEM Objects Creation
135--------------------
136
137GEM splits creation of GEM objects and allocation of the memory that
138backs them in two distinct operations.
139
140GEM objects are represented by an instance of struct :c:type:`struct
141drm_gem_object <drm_gem_object>`. Drivers usually need to
142extend GEM objects with private information and thus create a
143driver-specific GEM object structure type that embeds an instance of
144struct :c:type:`struct drm_gem_object <drm_gem_object>`.
145
146To create a GEM object, a driver allocates memory for an instance of its
147specific GEM object type and initializes the embedded struct
148:c:type:`struct drm_gem_object <drm_gem_object>` with a call
149to drm_gem_object_init(). The function takes a pointer
150to the DRM device, a pointer to the GEM object and the buffer object
151size in bytes.
152
153GEM uses shmem to allocate anonymous pageable memory.
154drm_gem_object_init() will create an shmfs file of the
155requested size and store it into the struct :c:type:`struct
156drm_gem_object <drm_gem_object>` filp field. The memory is
157used as either main storage for the object when the graphics hardware
158uses system memory directly or as a backing store otherwise. Drivers
159can call drm_gem_huge_mnt_create() to create, mount and use a huge
160shmem mountpoint instead of the default one ('shm_mnt'). For builds
161with CONFIG_TRANSPARENT_HUGEPAGE enabled, further calls to
162drm_gem_object_init() will let shmem allocate huge pages when
163possible.
164
165Drivers are responsible for the actual physical pages allocation by
166calling shmem_read_mapping_page_gfp() for each page.
167Note that they can decide to allocate pages when initializing the GEM
168object, or to delay allocation until the memory is needed (for instance
169when a page fault occurs as a result of a userspace memory access or
170when the driver needs to start a DMA transfer involving the memory).
171
172Anonymous pageable memory allocation is not always desired, for instance
173when the hardware requires physically contiguous system memory as is
174often the case in embedded devices. Drivers can create GEM objects with
175no shmfs backing (called private GEM objects) by initializing them with a call
176to drm_gem_private_object_init() instead of drm_gem_object_init(). Storage for
177private GEM objects must be managed by drivers.
178
179GEM Objects Lifetime
180--------------------
181
182All GEM objects are reference-counted by the GEM core. References can be
183acquired and release by calling drm_gem_object_get() and drm_gem_object_put()
184respectively.
185
186When the last reference to a GEM object is released the GEM core calls
187the :c:type:`struct drm_gem_object_funcs <gem_object_funcs>` free
188operation. That operation is mandatory for GEM-enabled drivers and must
189free the GEM object and all associated resources.
190
191void (\*free) (struct drm_gem_object \*obj); Drivers are
192responsible for freeing all GEM object resources. This includes the
193resources created by the GEM core, which need to be released with
194drm_gem_object_release().
195
196GEM Objects Naming
197------------------
198
199Communication between userspace and the kernel refers to GEM objects
200using local handles, global names or, more recently, file descriptors.
201All of those are 32-bit integer values; the usual Linux kernel limits
202apply to the file descriptors.
203
204GEM handles are local to a DRM file. Applications get a handle to a GEM
205object through a driver-specific ioctl, and can use that handle to refer
206to the GEM object in other standard or driver-specific ioctls. Closing a
207DRM file handle frees all its GEM handles and dereferences the
208associated GEM objects.
209
210To create a handle for a GEM object drivers call drm_gem_handle_create(). The
211function takes a pointer to the DRM file and the GEM object and returns a
212locally unique handle.  When the handle is no longer needed drivers delete it
213with a call to drm_gem_handle_delete(). Finally the GEM object associated with a
214handle can be retrieved by a call to drm_gem_object_lookup().
215
216Handles don't take ownership of GEM objects, they only take a reference
217to the object that will be dropped when the handle is destroyed. To
218avoid leaking GEM objects, drivers must make sure they drop the
219reference(s) they own (such as the initial reference taken at object
220creation time) as appropriate, without any special consideration for the
221handle. For example, in the particular case of combined GEM object and
222handle creation in the implementation of the dumb_create operation,
223drivers must drop the initial reference to the GEM object before
224returning the handle.
225
226GEM names are similar in purpose to handles but are not local to DRM
227files. They can be passed between processes to reference a GEM object
228globally. Names can't be used directly to refer to objects in the DRM
229API, applications must convert handles to names and names to handles
230using the DRM_IOCTL_GEM_FLINK and DRM_IOCTL_GEM_OPEN ioctls
231respectively. The conversion is handled by the DRM core without any
232driver-specific support.
233
234GEM also supports buffer sharing with dma-buf file descriptors through
235PRIME. GEM-based drivers must use the provided helpers functions to
236implement the exporting and importing correctly. See ?. Since sharing
237file descriptors is inherently more secure than the easily guessable and
238global GEM names it is the preferred buffer sharing mechanism. Sharing
239buffers through GEM names is only supported for legacy userspace.
240Furthermore PRIME also allows cross-device buffer sharing since it is
241based on dma-bufs.
242
243GEM Objects Mapping
244-------------------
245
246Because mapping operations are fairly heavyweight GEM favours
247read/write-like access to buffers, implemented through driver-specific
248ioctls, over mapping buffers to userspace. However, when random access
249to the buffer is needed (to perform software rendering for instance),
250direct access to the object can be more efficient.
251
252The mmap system call can't be used directly to map GEM objects, as they
253don't have their own file handle. Two alternative methods currently
254co-exist to map GEM objects to userspace. The first method uses a
255driver-specific ioctl to perform the mapping operation, calling
256do_mmap() under the hood. This is often considered
257dubious, seems to be discouraged for new GEM-enabled drivers, and will
258thus not be described here.
259
260The second method uses the mmap system call on the DRM file handle. void
261\*mmap(void \*addr, size_t length, int prot, int flags, int fd, off_t
262offset); DRM identifies the GEM object to be mapped by a fake offset
263passed through the mmap offset argument. Prior to being mapped, a GEM
264object must thus be associated with a fake offset. To do so, drivers
265must call drm_gem_create_mmap_offset() on the object.
266
267Once allocated, the fake offset value must be passed to the application
268in a driver-specific way and can then be used as the mmap offset
269argument.
270
271The GEM core provides a helper method drm_gem_mmap() to
272handle object mapping. The method can be set directly as the mmap file
273operation handler. It will look up the GEM object based on the offset
274value and set the VMA operations to the :c:type:`struct drm_driver
275<drm_driver>` gem_vm_ops field. Note that drm_gem_mmap() doesn't map memory to
276userspace, but relies on the driver-provided fault handler to map pages
277individually.
278
279To use drm_gem_mmap(), drivers must fill the struct :c:type:`struct drm_driver
280<drm_driver>` gem_vm_ops field with a pointer to VM operations.
281
282The VM operations is a :c:type:`struct vm_operations_struct <vm_operations_struct>`
283made up of several fields, the more interesting ones being:
284
285.. code-block:: c
286
287	struct vm_operations_struct {
288		void (*open)(struct vm_area_struct * area);
289		void (*close)(struct vm_area_struct * area);
290		vm_fault_t (*fault)(struct vm_fault *vmf);
291	};
292
293
294The open and close operations must update the GEM object reference
295count. Drivers can use the drm_gem_vm_open() and drm_gem_vm_close() helper
296functions directly as open and close handlers.
297
298The fault operation handler is responsible for mapping pages to
299userspace when a page fault occurs. Depending on the memory allocation
300scheme, drivers can allocate pages at fault time, or can decide to
301allocate memory for the GEM object at the time the object is created.
302
303Drivers that want to map the GEM object upfront instead of handling page
304faults can implement their own mmap file operation handler.
305
306In order to reduce page table overhead, if the internal shmem mountpoint
307"shm_mnt" is configured to use transparent huge pages (for builds with
308CONFIG_TRANSPARENT_HUGEPAGE enabled) and if the shmem backing store
309managed to allocate a huge page for a faulty address, the fault handler
310will first attempt to insert that huge page into the VMA before falling
311back to individual page insertion. mmap() user address alignment for GEM
312objects is handled by providing a custom get_unmapped_area file
313operation which forwards to the shmem backing store. For most drivers,
314which don't create a huge mountpoint by default or through a module
315parameter, transparent huge pages can be enabled by either setting the
316"transparent_hugepage_shmem" kernel parameter or the
317"/sys/kernel/mm/transparent_hugepage/shmem_enabled" sysfs knob.
318
319For platforms without MMU the GEM core provides a helper method
320drm_gem_dma_get_unmapped_area(). The mmap() routines will call this to get a
321proposed address for the mapping.
322
323To use drm_gem_dma_get_unmapped_area(), drivers must fill the struct
324:c:type:`struct file_operations <file_operations>` get_unmapped_area field with
325a pointer on drm_gem_dma_get_unmapped_area().
326
327More detailed information about get_unmapped_area can be found in
328Documentation/admin-guide/mm/nommu-mmap.rst
329
330Memory Coherency
331----------------
332
333When mapped to the device or used in a command buffer, backing pages for
334an object are flushed to memory and marked write combined so as to be
335coherent with the GPU. Likewise, if the CPU accesses an object after the
336GPU has finished rendering to the object, then the object must be made
337coherent with the CPU's view of memory, usually involving GPU cache
338flushing of various kinds. This core CPU<->GPU coherency management is
339provided by a device-specific ioctl, which evaluates an object's current
340domain and performs any necessary flushing or synchronization to put the
341object into the desired coherency domain (note that the object may be
342busy, i.e. an active render target; in that case, setting the domain
343blocks the client and waits for rendering to complete before performing
344any necessary flushing operations).
345
346Command Execution
347-----------------
348
349Perhaps the most important GEM function for GPU devices is providing a
350command execution interface to clients. Client programs construct
351command buffers containing references to previously allocated memory
352objects, and then submit them to GEM. At that point, GEM takes care to
353bind all the objects into the GTT, execute the buffer, and provide
354necessary synchronization between clients accessing the same buffers.
355This often involves evicting some objects from the GTT and re-binding
356others (a fairly expensive operation), and providing relocation support
357which hides fixed GTT offsets from clients. Clients must take care not
358to submit command buffers that reference more objects than can fit in
359the GTT; otherwise, GEM will reject them and no rendering will occur.
360Similarly, if several objects in the buffer require fence registers to
361be allocated for correct rendering (e.g. 2D blits on pre-965 chips),
362care must be taken not to require more fence registers than are
363available to the client. Such resource management should be abstracted
364from the client in libdrm.
365
366GEM Function Reference
367----------------------
368
369.. kernel-doc:: include/drm/drm_gem.h
370   :internal:
371
372.. kernel-doc:: drivers/gpu/drm/drm_gem.c
373   :export:
374
375GEM DMA Helper Functions Reference
376----------------------------------
377
378.. kernel-doc:: drivers/gpu/drm/drm_gem_dma_helper.c
379   :doc: dma helpers
380
381.. kernel-doc:: include/drm/drm_gem_dma_helper.h
382   :internal:
383
384.. kernel-doc:: drivers/gpu/drm/drm_gem_dma_helper.c
385   :export:
386
387GEM SHMEM Helper Function Reference
388-----------------------------------
389
390.. kernel-doc:: drivers/gpu/drm/drm_gem_shmem_helper.c
391   :doc: overview
392
393.. kernel-doc:: include/drm/drm_gem_shmem_helper.h
394   :internal:
395
396.. kernel-doc:: drivers/gpu/drm/drm_gem_shmem_helper.c
397   :export:
398
399GEM VRAM Helper Functions Reference
400-----------------------------------
401
402.. kernel-doc:: drivers/gpu/drm/drm_gem_vram_helper.c
403   :doc: overview
404
405.. kernel-doc:: include/drm/drm_gem_vram_helper.h
406   :internal:
407
408.. kernel-doc:: drivers/gpu/drm/drm_gem_vram_helper.c
409   :export:
410
411GEM TTM Helper Functions Reference
412-----------------------------------
413
414.. kernel-doc:: drivers/gpu/drm/drm_gem_ttm_helper.c
415   :doc: overview
416
417.. kernel-doc:: drivers/gpu/drm/drm_gem_ttm_helper.c
418   :export:
419
420VMA Offset Manager
421==================
422
423.. kernel-doc:: drivers/gpu/drm/drm_vma_manager.c
424   :doc: vma offset manager
425
426.. kernel-doc:: include/drm/drm_vma_manager.h
427   :internal:
428
429.. kernel-doc:: drivers/gpu/drm/drm_vma_manager.c
430   :export:
431
432.. _prime_buffer_sharing:
433
434PRIME Buffer Sharing
435====================
436
437PRIME is the cross device buffer sharing framework in drm, originally
438created for the OPTIMUS range of multi-gpu platforms. To userspace PRIME
439buffers are dma-buf based file descriptors.
440
441Overview and Lifetime Rules
442---------------------------
443
444.. kernel-doc:: drivers/gpu/drm/drm_prime.c
445   :doc: overview and lifetime rules
446
447PRIME Helper Functions
448----------------------
449
450.. kernel-doc:: drivers/gpu/drm/drm_prime.c
451   :doc: PRIME Helpers
452
453PRIME Function References
454-------------------------
455
456.. kernel-doc:: include/drm/drm_prime.h
457   :internal:
458
459.. kernel-doc:: drivers/gpu/drm/drm_prime.c
460   :export:
461
462DRM MM Range Allocator
463======================
464
465Overview
466--------
467
468.. kernel-doc:: drivers/gpu/drm/drm_mm.c
469   :doc: Overview
470
471LRU Scan/Eviction Support
472-------------------------
473
474.. kernel-doc:: drivers/gpu/drm/drm_mm.c
475   :doc: lru scan roster
476
477DRM MM Range Allocator Function References
478------------------------------------------
479
480.. kernel-doc:: include/drm/drm_mm.h
481   :internal:
482
483.. kernel-doc:: drivers/gpu/drm/drm_mm.c
484   :export:
485
486.. _drm_gpuvm:
487
488DRM GPUVM
489=========
490
491Overview
492--------
493
494.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c
495   :doc: Overview
496
497Split and Merge
498---------------
499
500.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c
501   :doc: Split and Merge
502
503.. _drm_gpuvm_locking:
504
505Locking
506-------
507
508.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c
509   :doc: Locking
510
511Examples
512--------
513
514.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c
515   :doc: Examples
516
517DRM GPUVM Function References
518-----------------------------
519
520.. kernel-doc:: include/drm/drm_gpuvm.h
521   :internal:
522
523.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c
524   :export:
525
526DRM Buddy Allocator
527===================
528
529DRM Buddy Function References
530-----------------------------
531
532.. kernel-doc:: drivers/gpu/drm/drm_buddy.c
533   :export:
534
535DRM Cache Handling and Fast WC memcpy()
536=======================================
537
538.. kernel-doc:: drivers/gpu/drm/drm_cache.c
539   :export:
540
541.. _drm_sync_objects:
542
543DRM Sync Objects
544================
545
546.. kernel-doc:: drivers/gpu/drm/drm_syncobj.c
547   :doc: Overview
548
549.. kernel-doc:: include/drm/drm_syncobj.h
550   :internal:
551
552.. kernel-doc:: drivers/gpu/drm/drm_syncobj.c
553   :export:
554
555DRM Execution context
556=====================
557
558.. kernel-doc:: drivers/gpu/drm/drm_exec.c
559   :doc: Overview
560
561.. kernel-doc:: include/drm/drm_exec.h
562   :internal:
563
564.. kernel-doc:: drivers/gpu/drm/drm_exec.c
565   :export:
566
567GPU Scheduler
568=============
569
570Overview
571--------
572
573.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c
574   :doc: Overview
575
576Flow Control
577------------
578
579.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c
580   :doc: Flow Control
581
582Scheduler Function References
583-----------------------------
584
585.. kernel-doc:: include/drm/gpu_scheduler.h
586   :internal:
587
588.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c
589   :export:
590
591.. kernel-doc:: drivers/gpu/drm/scheduler/sched_entity.c
592   :export:
593