The DRI3 Extension Version 1.3 2021-11-30 Keith Packard keithp@keithp.com Intel Corporation Daniel Stone daniels@collabora.com Collabora 1. Introduction The DRI3 extension provides mechanisms to translate between direct rendered buffers and X pixmaps. When combined with the Present extension, a complete direct rendering solution for hardware-accelerated devices such as GPUs is provided. The direct rendered buffers are passed across the protocol via standard POSIX file descriptor passing mechanisms. On Linux, these buffers are DMA-BUF objects. DRI3 also includes a mechanism to translate between Linux Futexes and X Sync extension Fences. This provides a synchronization mechanism which can be used to serialize access to shared render buffers. 1.1. Acknowledgments Eric Anholt Dave Airlie Kristian Høgsberg James Jones Arthur Huillet Louis-Francis Ratté-Boulianne Austin Shafer ❄ ❄ ❄ ❄ ❄ ❄ ❄ 2. Data Types The DRI3 extension uses the RandR extension Provider data type to select among multiple GPUs on a single screen and the Sync extension fence object to provide graphics object synchronization. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 4. Errors DRI3 defines no errors. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 5. Events DRI3 defines no events. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 6. Protocol Types DRI3 defines no new protocol types. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 7. Extension Initialization The name of this extension is "DRI3" ┌─── DRI3QueryVersion client-major-version: CARD32 client-minor-version: CARD32 ▶ major-version: CARD32 minor-version: CARD32 └─── The client sends the highest supported version to the server and the server sends the highest version it supports, but no higher than the requested version. Major versions changes can introduce incompatibilities in existing functionality, minor version changes introduce only backward compatible changes. It is the clients responsibility to ensure that the server supports a version which is compatible with its expectations. Backwards compatible changes included addition of new requests. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 8. Extension Requests ┌─── DRI3Open drawable: DRAWABLE provider: PROVIDER ▶ nfd: CARD8 device: FD └─── Errors: Drawable, Value, Match This requests that the X server open the direct rendering device associated with drawable and RandR provider. The provider must support SourceOutput or SourceOffload. The file descriptor for the device is returned in 'device'. 'nfd' will be set to one (this is strictly a convenience for XCB which otherwise would need request-specific information about how many file descriptors were associated with this reply). ┌─── DRI3PixmapFromBuffer pixmap: PIXMAP drawable: DRAWABLE size: CARD32 width, height, stride: CARD16 depth, bpp: CARD8 buffer: FD └─── Errors: Alloc, Drawable, IDChoice, Value, Match Creates a pixmap for the direct rendering object associated with 'buffer' and the screen associated with 'drawable'. Changes to pixmap will be visible in that direct rendered object and changes to the direct rendered object will be visible in the pixmap. 'size' specifies the total size of the buffer bytes. 'width', 'height' describe the geometry (in pixels) of the underlying buffer. 'stride' specifies the number of bytes per scanline in the buffer. The pixels within the buffer are not required to be arranged in a simple linear fashion, but 'size' will be at least 'height' * 'stride'. Precisely how any additional information about the buffer is shared is outside the scope of this extension. If buffer cannot be used with the screen associated with drawable, a Match error is returned. If depth or bpp are not supported by the screen, a Value error is returned. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3BufferFromPixmap pixmap: PIXMAP ▶ nfd: CARD8 size: CARD32 width, height, stride: CARD16 depth, bpp: CARD8 buffer: FD └─── Errors: Pixmap, Match Pass back a direct rendering object associated with pixmap. Changes to pixmap will be visible in that direct rendered object and changes to the direct rendered object will be visible in the pixmap. 'size' specifies the total size of the buffer bytes. 'width', 'height' describe the geometry (in pixels) of the underlying buffer. 'stride' specifies the number of bytes per scanline in the buffer. The pixels within the buffer are not required to be arranged in a simple linear fashion, but 'size' will be at least 'height' * 'stride'. Precisely how any additional information about the buffer is shared is outside the scope of this extension. If buffer cannot be used with the screen associated with drawable, a Match error is returned. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3FenceFromFD drawable: DRAWABLE fence: FENCE initially-triggered: BOOL fd: FD └─── Errors: IDchoice, Drawable Creates a Sync extension Fence that provides the regular Sync extension semantics along with a file descriptor that provides a device-specific mechanism to manipulate the fence directly. Details about the mechanism used with this file descriptor are outside the scope of the DRI3 extension. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3FDFromFence drawable: DRAWABLE fence: FENCE ▶ fd: FD └─── Errors: IDchoice, Drawable, Match Given a Sync extension Fence that provides the regular Sync extension semantics, returns a file descriptor that provides a device-specific mechanism to manipulate the fence directly. Details about the mechanism used with this file descriptor are outside the scope of the DRI3 extension. 'drawable' must be associated with a direct rendering device that 'fence' can work with, otherwise a Match error results. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3GetSupportedModifiers window: WINDOW depth: CARD8 bpp: CARD8 ▶ num_window_modifiers: CARD32 num_screen_modifiers: CARD32 window_modifiers: ListOfCARD64 screen_modifiers: ListOfCARD64 └─── Errors: Window, Match Return supported DRM FourCC modifiers for the specified 'window'. The first list of 'window_modifiers' contains a set of modifiers which the server considers optimal for the window's current configuration. Using these modifiers to allocate, even if locally suboptimal to the client driver, may result in a more optimal display pipeline, e.g. by avoiding composition. The second list of 'screen_modifiers', is the total set of modifiers which are acceptable for use on the Screen associated with 'window'. This set of modifiers will not change over the lifetime of the client. Using this set of modifiers to allocate may not result in a globally optimal pipeline, if separate 'window_modifiers' are available. It is expected that a client calling this request will obtain the modifiers for a particular window, allocate buffers using the preferred modifier set as described above, create a Pixmap referring to the storage of those buffers using the DRI3BuffersFromPixmap request, then make the content visible in the storage of those buffers visible with a request such as the Present extension's PresentPixmap. The meaning of any modifier is canonically defined in drm_fourcc.h. ┌─── DRI3PixmapFromBuffers pixmap: PIXMAP window: WINDOW num_buffers: CARD8 width, height: CARD16 stride0, offset0: CARD32 stride1, offset1: CARD32 stride2, offset2: CARD32 stride3, offset3: CARD32 depth, bpp: CARD8 modifier: CARD64 buffers: ListOfFD └─── Errors: Alloc, Window, IDChoice, Value, Match Creates a pixmap for the direct rendering object associated with 'buffers' and the screen associated with 'window'. Changes to pixmap will be visible in that direct rendered object and changes to the direct rendered object will be visible in the pixmap. The pixmap will be available for presentation to the window. In contrast to PixmapFromBuffer, multiple buffers may be combined to specify a single logical source for pixel sampling: 'num_buffers' may be set from 1 (single buffer, akin to PixmapFromBuffer) to 4. This is the number of file descriptors which will be sent with this request; one per buffer. Modifiers allow explicit specification of non-linear sources, such as tiled or compressed buffers. The combination of bpp, depth, and modifier allows unambiguous declaration of the buffer layout in a manner defined by the DRM tokens. If 'modifier' is DRM_FORMAT_MOD_INVALID, the client does not have information on the buffer layout. In this case, the buffer may only have a single plane. The driver may make its own inference through unspecified means to determine the exact buffer layout, however this is neither required nor defined by the specification, and is considered an implementation detail of the particular driver. 'width' and 'height' describe the geometry (in pixels) of the logical pixel-sample source. 'strideN' and 'offsetN' define the number of bytes per logical scanline, and the distance in bytes from the beginning of the buffer passed for that plane until the start of the sample source for that plane, respectively for plane N. If the plane is not used according to the format and modifier specification, both values for that plane must be zero. Precisely how any additional information about the buffer (such as memory placement) is shared is outside the scope of this extension. If the buffer(s) cannot be used with the screen associated with 'window', a Match error is returned. If the bpp, depth, and modifier combination is not supported by the screen, a Value error is returned. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3BuffersFromPixmap pixmap: PIXMAP ▶ nfd: CARD8 width, height: CARD16 depth, bpp: CARD8 modifier: CARD64 strides: ListOfCARD32 offsets: ListOfCARD32 buffers: ListOfFD └─── Errors: Pixmap, Match Returns direct rendering objects associated with 'pixmap'. Changes to 'pixmap' will be visible in the direct rendered objects and changes to the direct rendered objects will be visible in 'pixmap' after flushing and synchronization. 'width' and 'height' describe the geometry (in pixels) of the logical pixel-sample source from combining the direct rendering objects. See PixmapFromBuffers for more details on DRM modifiers usage. 'nfd' describes the number of buffers returned for the pixmap, which must be combined together according to 'depth', 'bpp', and 'modifier'. For each buffer, there is an entry in the 'strides', 'offsets', and 'buffers' list. 'buffer' contains a single file descriptor referring to the buffer. 'stride' specifies the number of bytes per logical scanline for this plane, and 'offset' specifies the distance in bytes from the beginning of 'buffer' until the start of the sample source for that plane. Precisely how any additional information about the buffer is shared is outside the scope of this extension. If buffers cannot be exported from the the screen associated with 'pixmap', a Match error is returned. For information on synchronization of buffer access between the client and the X server, please see section 12. ┌─── DRI3SetDRMDeviceInUse window: WINDOW drmMajor: CARD32 drmMinor: CARD32 └─── Errors: Window This request provides a hint to the server about the device in use by this window. This is used to provide DRI3GetSupportedModifiers with a hint of what device to return modifiers for in the window_modifiers return value. Using this hint allows for device-specific modifiers to be returned by DRI3GetSupportedModifiers, for example when an application is renderoffloaded and eligible for direct scanout. This is only a hint, and may or may not be reflected in the modifier list returned. If the window specified was not found, a Window error will be returned. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 9. Extension Events DRI3 defines no events. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 10. Extension Versioning The DRI3 extension is adapted from the DRI2 extension. 1.0: First published version 1.1: Cosmetic changes 1.2: Add GetSupportedModifiers, PixmapFromBuffers, and BuffersFromPixmap requests. 1.3: Add SetDRMDeviceInUse ❄ ❄ ❄ ❄ ❄ ❄ ❄ 11. Relationship with other extensions As an extension designed to support other extensions, there is naturally some interactions with other extensions. 11.1 GLX GLX is both an application interface and an X extension. OpenGL applications using the GLX API will use the GLX extension, DRI3 and Present when doing direct rendering. 11.2 Present The Present extension provides a way to synchronize the display of pixmap contents to the screen. When used in conjunction with DRI3, they provide a complete direct rendering solution for OpenGL or other APIs. 11.3 DRI2 DRI3 provides similar functionality to the DRI2Connect and DRI2GetBuffersWithFormat requests, however DRI3 uses file descriptors to refer to the direct rendering device and buffers. Present and DRI3 are designed in conjunction to replace DRI2 11.2 XvMC / Xv It might be nice to be able to reference YUV formatted direct rendered objects from the X server. ❄ ❄ ❄ ❄ ❄ ❄ ❄ 12. Synchronization Synchronization of access to buffers shared between processes is not currently explicitly controlled by this protocol. Without the use of additional extensions not defined by the DRI3 protocol as of version 1.2, synchronization between multiple processes and contexts is considered to follow the implicit model. In this model, the driver is required to have a global view of access requests issued by all processes with a reference to the buffer, and control scheduling of all operations on that buffer, whether performed by the CPU or auxiliary hardware. The driver is responsible for enforcing a strict ordering to protect against write-after-read or read-after-write hazards, such that any reads requested by one process or context, are fulfilled before any writes requested by another process or context, as long as that read was definitively submitted before the write. A similar dependency exists for reads submitted after writes: the driver must ensure that the write is fully visible and coherent to the read request. As a purely illustrative example, if two processes share a buffer, where one process reads from a buffer using an OpenGL texture sampler and submits this work by calling 'glFlush', and the other process submits work to the driver to write to that buffer, the driver is responsible for ensuring that the results of the latter write are not visible to the texture sampler. The Sync fences provided by DRI3 control only this submission of work and ensuing global visibility of the requests, rather than the completion of the work within any hardware. To further the example above, a fence used to prevent any writes to the buffer before the sampler had completed access, the fence would be signaled when 'glFlush' had been called, at which point the request has become globally visible to the driver's request-scheduling and synchronization mechanisms. The logical ordering of requests made by software has been preserved, and the driver then takes care to ensure that these requests are scheduled such they do not observe effects from requests made later in time. This presents a fully coherent in-order FIFO-like model across processes, where synchronzation is handled externally to the DRI3 client with no explicit intervention. This restriction also applies for cross-device usage. ❄ ❄ ❄ ❄ ❄ ❄ ❄ Appendix A. Protocol Encoding Syntactic Conventions This document uses the same syntactic conventions as the core X protocol encoding document. A.1 Common Types None. A.2 Protocol Requests ┌─── DRI3QueryVersion 1 CARD8 major opcode 1 0 DRI3 opcode 2 3 length 4 CARD32 major version 4 CARD32 minor version ▶ 1 1 Reply 1 unused 2 CARD16 sequence number 4 0 reply length 4 CARD32 major version 4 CARD32 minor version 16 unused └─── ┌─── DRI3Open 1 CARD8 major opcode 1 1 DRI3 opcode 2 4 length 4 DRAWABLE drawable 4 PROVIDER provider ▶ 1 1 Reply 1 1 nfd 2 CARD16 sequence number 4 0 reply length 24 unused 0 FD device └─── ┌─── DRI3PixmapFromBuffer 1 CARD8 major opcode 1 2 DRI3 opcode 2 6 length 4 Pixmap pixmap 4 Drawable drawable 4 CARD32 size 2 CARD16 width 2 CARD16 height 2 CARD16 stride 1 CARD8 depth 1 CARD8 bpp 0 FD buffer └─── ┌─── DRI3BufferFromPixmap 1 CARD8 major opcode 1 3 DRI3 opcode 2 2 length 4 Pixmap pixmap ▶ 1 1 Reply 1 1 nfd 2 CARD16 sequence number 4 0 reply length 4 CARD32 size 2 CARD16 width 2 CARD16 height 2 CARD16 stride 1 CARD8 depth 1 CARD8 bpp 12 unused 0 FD buffer └─── ┌─── DRI3FenceFromFD 1 CARD8 major opcode 1 4 DRI3 opcode 2 4 length 4 Drawable drawable 4 Fence fence 1 BOOL initially triggered 3 unused 0 FD fence fd └─── ┌─── DRI3FDFromFence 1 CARD8 major opcode 1 5 DRI3 opcode 2 3 length 4 Drawable drawable 4 Fence fence ▶ 1 1 Reply 1 1 nfd 2 CARD16 sequence number 4 0 reply length 24 unused 0 FD fence fd └─── ┌─── DRI3GetSupportedModifiers 1 CARD8 major opcode 1 7 DRI3 opcode 2 3 length 4 Window window 1 CARD8 depth 1 CARD8 bpp 2 unused ▶ 1 1 Reply 1 0 unused 2 CARD16 sequence number 4 CARD32 reply length 4 CARD32 num_window_modifiers 4 CARD32 num_screen_modifiers 16 unused 4 ListOfCARD64 window_modifiers[num_window_modifiers] 4 ListOfCARD64 screen_modifiers[num_screen_modifiers] └─── ┌─── DRI3PixmapFromBuffers 1 CARD8 major opcode 1 8 DRI3 opcode 2 8 length 4 Pixmap pixmap 4 Window window 1 CARD8 num_buffers 3 unused 2 CARD16 width 2 CARD16 height 4 CARD32 stride0 4 CARD32 offset0 4 CARD32 stride1 4 CARD32 offset1 4 CARD32 stride2 4 CARD32 offset2 4 CARD32 stride3 4 CARD32 offset3 1 CARD8 depth 1 CARD8 bpp 2 unused 8 CARD64 modifier 0 ListOfFD buffers[num_buffers] └─── ┌─── DRI3BuffersFromPixmap 1 CARD8 major opcode 1 9 DRI3 opcode 2 2 length 4 Pixmap pixmap ▶ 1 1 Reply 1 CARD8 nfd 2 CARD16 sequence number 4 CARD32 reply length 2 CARD16 width 2 CARD16 height 4 CARD8 unused 8 CARD64 modifier 1 CARD8 depth 1 CARD8 bpp 6 unused 0 ListOfFD buffer[nfd] 4 ListOfCARD32 strides[nfd] 4 ListOfCARD32 offsets[nfd] └─── ┌─── DRI3SetDRMDeviceInUse 1 CARD8 major opcode 1 10 DRI3 opcode 2 4 length 4 Window window 4 CARD32 drmMajor 4 CARD32 drmMinor └─── A.3 Protocol Events The DRI3 extension defines no events. A.4 Protocol Errors The DRI3 extension defines no errors. ❄ ❄ ❄ ❄ ❄ ❄ ❄