576 lines · plain
1=====================2DRM Memory Management3=====================4 5Modern Linux systems require large amount of graphics memory to store6frame buffers, textures, vertices and other graphics-related data. Given7the very dynamic nature of many of that data, managing graphics memory8efficiently is thus crucial for the graphics stack and plays a central9role in the DRM infrastructure.10 11The DRM core includes two memory managers, namely Translation Table Manager12(TTM) and Graphics Execution Manager (GEM). TTM was the first DRM memory13manager to be developed and tried to be a one-size-fits-them all14solution. It provides a single userspace API to accommodate the need of15all hardware, supporting both Unified Memory Architecture (UMA) devices16and devices with dedicated video RAM (i.e. most discrete video cards).17This resulted in a large, complex piece of code that turned out to be18hard to use for driver development.19 20GEM started as an Intel-sponsored project in reaction to TTM's21complexity. Its design philosophy is completely different: instead of22providing a solution to every graphics memory-related problems, GEM23identified common code between drivers and created a support library to24share it. GEM has simpler initialization and execution requirements than25TTM, but has no video RAM management capabilities and is thus limited to26UMA devices.27 28The Translation Table Manager (TTM)29===================================30 31.. kernel-doc:: drivers/gpu/drm/ttm/ttm_module.c32 :doc: TTM33 34.. kernel-doc:: include/drm/ttm/ttm_caching.h35 :internal:36 37TTM device object reference38---------------------------39 40.. kernel-doc:: include/drm/ttm/ttm_device.h41 :internal:42 43.. kernel-doc:: drivers/gpu/drm/ttm/ttm_device.c44 :export:45 46TTM resource placement reference47--------------------------------48 49.. kernel-doc:: include/drm/ttm/ttm_placement.h50 :internal:51 52TTM resource object reference53-----------------------------54 55.. kernel-doc:: include/drm/ttm/ttm_resource.h56 :internal:57 58.. kernel-doc:: drivers/gpu/drm/ttm/ttm_resource.c59 :export:60 61TTM TT object reference62-----------------------63 64.. kernel-doc:: include/drm/ttm/ttm_tt.h65 :internal:66 67.. kernel-doc:: drivers/gpu/drm/ttm/ttm_tt.c68 :export:69 70TTM page pool reference71-----------------------72 73.. kernel-doc:: include/drm/ttm/ttm_pool.h74 :internal:75 76.. kernel-doc:: drivers/gpu/drm/ttm/ttm_pool.c77 :export:78 79The Graphics Execution Manager (GEM)80====================================81 82The GEM design approach has resulted in a memory manager that doesn't83provide full coverage of all (or even all common) use cases in its84userspace or kernel API. GEM exposes a set of standard memory-related85operations to userspace and a set of helper functions to drivers, and86let drivers implement hardware-specific operations with their own87private API.88 89The GEM userspace API is described in the `GEM - the Graphics Execution90Manager <http://lwn.net/Articles/283798/>`__ article on LWN. While91slightly outdated, the document provides a good overview of the GEM API92principles. Buffer allocation and read and write operations, described93as part of the common GEM API, are currently implemented using94driver-specific ioctls.95 96GEM is data-agnostic. It manages abstract buffer objects without knowing97what individual buffers contain. APIs that require knowledge of buffer98contents or purpose, such as buffer allocation or synchronization99primitives, are thus outside of the scope of GEM and must be implemented100using driver-specific ioctls.101 102On a fundamental level, GEM involves several operations:103 104- Memory allocation and freeing105- Command execution106- Aperture management at command execution time107 108Buffer object allocation is relatively straightforward and largely109provided by Linux's shmem layer, which provides memory to back each110object.111 112Device-specific operations, such as command execution, pinning, buffer113read & write, mapping, and domain ownership transfers are left to114driver-specific ioctls.115 116GEM Initialization117------------------118 119Drivers that use GEM must set the DRIVER_GEM bit in the struct120:c:type:`struct drm_driver <drm_driver>` driver_features121field. The DRM core will then automatically initialize the GEM core122before calling the load operation. Behind the scene, this will create a123DRM Memory Manager object which provides an address space pool for124object allocation.125 126In a KMS configuration, drivers need to allocate and initialize a127command ring buffer following core GEM initialization if required by the128hardware. UMA devices usually have what is called a "stolen" memory129region, which provides space for the initial framebuffer and large,130contiguous memory regions required by the device. This space is131typically not managed by GEM, and must be initialized separately into132its own DRM MM object.133 134GEM Objects Creation135--------------------136 137GEM splits creation of GEM objects and allocation of the memory that138backs them in two distinct operations.139 140GEM objects are represented by an instance of struct :c:type:`struct141drm_gem_object <drm_gem_object>`. Drivers usually need to142extend GEM objects with private information and thus create a143driver-specific GEM object structure type that embeds an instance of144struct :c:type:`struct drm_gem_object <drm_gem_object>`.145 146To create a GEM object, a driver allocates memory for an instance of its147specific GEM object type and initializes the embedded struct148:c:type:`struct drm_gem_object <drm_gem_object>` with a call149to drm_gem_object_init(). The function takes a pointer150to the DRM device, a pointer to the GEM object and the buffer object151size in bytes.152 153GEM uses shmem to allocate anonymous pageable memory.154drm_gem_object_init() will create an shmfs file of the155requested size and store it into the struct :c:type:`struct156drm_gem_object <drm_gem_object>` filp field. The memory is157used as either main storage for the object when the graphics hardware158uses system memory directly or as a backing store otherwise.159 160Drivers are responsible for the actual physical pages allocation by161calling shmem_read_mapping_page_gfp() for each page.162Note that they can decide to allocate pages when initializing the GEM163object, or to delay allocation until the memory is needed (for instance164when a page fault occurs as a result of a userspace memory access or165when the driver needs to start a DMA transfer involving the memory).166 167Anonymous pageable memory allocation is not always desired, for instance168when the hardware requires physically contiguous system memory as is169often the case in embedded devices. Drivers can create GEM objects with170no shmfs backing (called private GEM objects) by initializing them with a call171to drm_gem_private_object_init() instead of drm_gem_object_init(). Storage for172private GEM objects must be managed by drivers.173 174GEM Objects Lifetime175--------------------176 177All GEM objects are reference-counted by the GEM core. References can be178acquired and release by calling drm_gem_object_get() and drm_gem_object_put()179respectively.180 181When the last reference to a GEM object is released the GEM core calls182the :c:type:`struct drm_gem_object_funcs <gem_object_funcs>` free183operation. That operation is mandatory for GEM-enabled drivers and must184free the GEM object and all associated resources.185 186void (\*free) (struct drm_gem_object \*obj); Drivers are187responsible for freeing all GEM object resources. This includes the188resources created by the GEM core, which need to be released with189drm_gem_object_release().190 191GEM Objects Naming192------------------193 194Communication between userspace and the kernel refers to GEM objects195using local handles, global names or, more recently, file descriptors.196All of those are 32-bit integer values; the usual Linux kernel limits197apply to the file descriptors.198 199GEM handles are local to a DRM file. Applications get a handle to a GEM200object through a driver-specific ioctl, and can use that handle to refer201to the GEM object in other standard or driver-specific ioctls. Closing a202DRM file handle frees all its GEM handles and dereferences the203associated GEM objects.204 205To create a handle for a GEM object drivers call drm_gem_handle_create(). The206function takes a pointer to the DRM file and the GEM object and returns a207locally unique handle. When the handle is no longer needed drivers delete it208with a call to drm_gem_handle_delete(). Finally the GEM object associated with a209handle can be retrieved by a call to drm_gem_object_lookup().210 211Handles don't take ownership of GEM objects, they only take a reference212to the object that will be dropped when the handle is destroyed. To213avoid leaking GEM objects, drivers must make sure they drop the214reference(s) they own (such as the initial reference taken at object215creation time) as appropriate, without any special consideration for the216handle. For example, in the particular case of combined GEM object and217handle creation in the implementation of the dumb_create operation,218drivers must drop the initial reference to the GEM object before219returning the handle.220 221GEM names are similar in purpose to handles but are not local to DRM222files. They can be passed between processes to reference a GEM object223globally. Names can't be used directly to refer to objects in the DRM224API, applications must convert handles to names and names to handles225using the DRM_IOCTL_GEM_FLINK and DRM_IOCTL_GEM_OPEN ioctls226respectively. The conversion is handled by the DRM core without any227driver-specific support.228 229GEM also supports buffer sharing with dma-buf file descriptors through230PRIME. GEM-based drivers must use the provided helpers functions to231implement the exporting and importing correctly. See ?. Since sharing232file descriptors is inherently more secure than the easily guessable and233global GEM names it is the preferred buffer sharing mechanism. Sharing234buffers through GEM names is only supported for legacy userspace.235Furthermore PRIME also allows cross-device buffer sharing since it is236based on dma-bufs.237 238GEM Objects Mapping239-------------------240 241Because mapping operations are fairly heavyweight GEM favours242read/write-like access to buffers, implemented through driver-specific243ioctls, over mapping buffers to userspace. However, when random access244to the buffer is needed (to perform software rendering for instance),245direct access to the object can be more efficient.246 247The mmap system call can't be used directly to map GEM objects, as they248don't have their own file handle. Two alternative methods currently249co-exist to map GEM objects to userspace. The first method uses a250driver-specific ioctl to perform the mapping operation, calling251do_mmap() under the hood. This is often considered252dubious, seems to be discouraged for new GEM-enabled drivers, and will253thus not be described here.254 255The second method uses the mmap system call on the DRM file handle. void256\*mmap(void \*addr, size_t length, int prot, int flags, int fd, off_t257offset); DRM identifies the GEM object to be mapped by a fake offset258passed through the mmap offset argument. Prior to being mapped, a GEM259object must thus be associated with a fake offset. To do so, drivers260must call drm_gem_create_mmap_offset() on the object.261 262Once allocated, the fake offset value must be passed to the application263in a driver-specific way and can then be used as the mmap offset264argument.265 266The GEM core provides a helper method drm_gem_mmap() to267handle object mapping. The method can be set directly as the mmap file268operation handler. It will look up the GEM object based on the offset269value and set the VMA operations to the :c:type:`struct drm_driver270<drm_driver>` gem_vm_ops field. Note that drm_gem_mmap() doesn't map memory to271userspace, but relies on the driver-provided fault handler to map pages272individually.273 274To use drm_gem_mmap(), drivers must fill the struct :c:type:`struct drm_driver275<drm_driver>` gem_vm_ops field with a pointer to VM operations.276 277The VM operations is a :c:type:`struct vm_operations_struct <vm_operations_struct>`278made up of several fields, the more interesting ones being:279 280.. code-block:: c281 282 struct vm_operations_struct {283 void (*open)(struct vm_area_struct * area);284 void (*close)(struct vm_area_struct * area);285 vm_fault_t (*fault)(struct vm_fault *vmf);286 };287 288 289The open and close operations must update the GEM object reference290count. Drivers can use the drm_gem_vm_open() and drm_gem_vm_close() helper291functions directly as open and close handlers.292 293The fault operation handler is responsible for mapping individual pages294to userspace when a page fault occurs. Depending on the memory295allocation scheme, drivers can allocate pages at fault time, or can296decide to allocate memory for the GEM object at the time the object is297created.298 299Drivers that want to map the GEM object upfront instead of handling page300faults can implement their own mmap file operation handler.301 302For platforms without MMU the GEM core provides a helper method303drm_gem_dma_get_unmapped_area(). The mmap() routines will call this to get a304proposed address for the mapping.305 306To use drm_gem_dma_get_unmapped_area(), drivers must fill the struct307:c:type:`struct file_operations <file_operations>` get_unmapped_area field with308a pointer on drm_gem_dma_get_unmapped_area().309 310More detailed information about get_unmapped_area can be found in311Documentation/admin-guide/mm/nommu-mmap.rst312 313Memory Coherency314----------------315 316When mapped to the device or used in a command buffer, backing pages for317an object are flushed to memory and marked write combined so as to be318coherent with the GPU. Likewise, if the CPU accesses an object after the319GPU has finished rendering to the object, then the object must be made320coherent with the CPU's view of memory, usually involving GPU cache321flushing of various kinds. This core CPU<->GPU coherency management is322provided by a device-specific ioctl, which evaluates an object's current323domain and performs any necessary flushing or synchronization to put the324object into the desired coherency domain (note that the object may be325busy, i.e. an active render target; in that case, setting the domain326blocks the client and waits for rendering to complete before performing327any necessary flushing operations).328 329Command Execution330-----------------331 332Perhaps the most important GEM function for GPU devices is providing a333command execution interface to clients. Client programs construct334command buffers containing references to previously allocated memory335objects, and then submit them to GEM. At that point, GEM takes care to336bind all the objects into the GTT, execute the buffer, and provide337necessary synchronization between clients accessing the same buffers.338This often involves evicting some objects from the GTT and re-binding339others (a fairly expensive operation), and providing relocation support340which hides fixed GTT offsets from clients. Clients must take care not341to submit command buffers that reference more objects than can fit in342the GTT; otherwise, GEM will reject them and no rendering will occur.343Similarly, if several objects in the buffer require fence registers to344be allocated for correct rendering (e.g. 2D blits on pre-965 chips),345care must be taken not to require more fence registers than are346available to the client. Such resource management should be abstracted347from the client in libdrm.348 349GEM Function Reference350----------------------351 352.. kernel-doc:: include/drm/drm_gem.h353 :internal:354 355.. kernel-doc:: drivers/gpu/drm/drm_gem.c356 :export:357 358GEM DMA Helper Functions Reference359----------------------------------360 361.. kernel-doc:: drivers/gpu/drm/drm_gem_dma_helper.c362 :doc: dma helpers363 364.. kernel-doc:: include/drm/drm_gem_dma_helper.h365 :internal:366 367.. kernel-doc:: drivers/gpu/drm/drm_gem_dma_helper.c368 :export:369 370GEM SHMEM Helper Function Reference371-----------------------------------372 373.. kernel-doc:: drivers/gpu/drm/drm_gem_shmem_helper.c374 :doc: overview375 376.. kernel-doc:: include/drm/drm_gem_shmem_helper.h377 :internal:378 379.. kernel-doc:: drivers/gpu/drm/drm_gem_shmem_helper.c380 :export:381 382GEM VRAM Helper Functions Reference383-----------------------------------384 385.. kernel-doc:: drivers/gpu/drm/drm_gem_vram_helper.c386 :doc: overview387 388.. kernel-doc:: include/drm/drm_gem_vram_helper.h389 :internal:390 391.. kernel-doc:: drivers/gpu/drm/drm_gem_vram_helper.c392 :export:393 394GEM TTM Helper Functions Reference395-----------------------------------396 397.. kernel-doc:: drivers/gpu/drm/drm_gem_ttm_helper.c398 :doc: overview399 400.. kernel-doc:: drivers/gpu/drm/drm_gem_ttm_helper.c401 :export:402 403VMA Offset Manager404==================405 406.. kernel-doc:: drivers/gpu/drm/drm_vma_manager.c407 :doc: vma offset manager408 409.. kernel-doc:: include/drm/drm_vma_manager.h410 :internal:411 412.. kernel-doc:: drivers/gpu/drm/drm_vma_manager.c413 :export:414 415.. _prime_buffer_sharing:416 417PRIME Buffer Sharing418====================419 420PRIME is the cross device buffer sharing framework in drm, originally421created for the OPTIMUS range of multi-gpu platforms. To userspace PRIME422buffers are dma-buf based file descriptors.423 424Overview and Lifetime Rules425---------------------------426 427.. kernel-doc:: drivers/gpu/drm/drm_prime.c428 :doc: overview and lifetime rules429 430PRIME Helper Functions431----------------------432 433.. kernel-doc:: drivers/gpu/drm/drm_prime.c434 :doc: PRIME Helpers435 436PRIME Function References437-------------------------438 439.. kernel-doc:: include/drm/drm_prime.h440 :internal:441 442.. kernel-doc:: drivers/gpu/drm/drm_prime.c443 :export:444 445DRM MM Range Allocator446======================447 448Overview449--------450 451.. kernel-doc:: drivers/gpu/drm/drm_mm.c452 :doc: Overview453 454LRU Scan/Eviction Support455-------------------------456 457.. kernel-doc:: drivers/gpu/drm/drm_mm.c458 :doc: lru scan roster459 460DRM MM Range Allocator Function References461------------------------------------------462 463.. kernel-doc:: include/drm/drm_mm.h464 :internal:465 466.. kernel-doc:: drivers/gpu/drm/drm_mm.c467 :export:468 469.. _drm_gpuvm:470 471DRM GPUVM472=========473 474Overview475--------476 477.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c478 :doc: Overview479 480Split and Merge481---------------482 483.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c484 :doc: Split and Merge485 486.. _drm_gpuvm_locking:487 488Locking489-------490 491.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c492 :doc: Locking493 494Examples495--------496 497.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c498 :doc: Examples499 500DRM GPUVM Function References501-----------------------------502 503.. kernel-doc:: include/drm/drm_gpuvm.h504 :internal:505 506.. kernel-doc:: drivers/gpu/drm/drm_gpuvm.c507 :export:508 509DRM Buddy Allocator510===================511 512DRM Buddy Function References513-----------------------------514 515.. kernel-doc:: drivers/gpu/drm/drm_buddy.c516 :export:517 518DRM Cache Handling and Fast WC memcpy()519=======================================520 521.. kernel-doc:: drivers/gpu/drm/drm_cache.c522 :export:523 524.. _drm_sync_objects:525 526DRM Sync Objects527================528 529.. kernel-doc:: drivers/gpu/drm/drm_syncobj.c530 :doc: Overview531 532.. kernel-doc:: include/drm/drm_syncobj.h533 :internal:534 535.. kernel-doc:: drivers/gpu/drm/drm_syncobj.c536 :export:537 538DRM Execution context539=====================540 541.. kernel-doc:: drivers/gpu/drm/drm_exec.c542 :doc: Overview543 544.. kernel-doc:: include/drm/drm_exec.h545 :internal:546 547.. kernel-doc:: drivers/gpu/drm/drm_exec.c548 :export:549 550GPU Scheduler551=============552 553Overview554--------555 556.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c557 :doc: Overview558 559Flow Control560------------561 562.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c563 :doc: Flow Control564 565Scheduler Function References566-----------------------------567 568.. kernel-doc:: include/drm/gpu_scheduler.h569 :internal:570 571.. kernel-doc:: drivers/gpu/drm/scheduler/sched_main.c572 :export:573 574.. kernel-doc:: drivers/gpu/drm/scheduler/sched_entity.c575 :export:576