1 .. Copyright 2020 DisplayLink (UK) Ltd.
7 The DRM core exports several interfaces to applications, generally
8 intended to be used through corresponding libdrm wrapper functions. In
9 addition, drivers export device-specific interfaces for use by userspace
10 drivers & device-aware applications through ioctls and sysfs files.
12 External interfaces include: memory mapping, context management, DMA
13 operations, AGP management, vblank control, fence management, memory
14 management, and output management.
16 Cover generic ioctls and sysfs layout here. We only need high-level
17 info, since man pages should cover the rest.
22 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
23 :doc: getunique and setversion story
28 Primary Nodes, DRM Master and Authentication
29 ============================================
31 .. kernel-doc:: drivers/gpu/drm/drm_auth.c
32 :doc: master and authentication
34 .. kernel-doc:: drivers/gpu/drm/drm_auth.c
37 .. kernel-doc:: include/drm/drm_auth.h
43 DRM Display Resource Leasing
44 ============================
46 .. kernel-doc:: drivers/gpu/drm/drm_lease.c
49 Open-Source Userspace Requirements
50 ==================================
52 The DRM subsystem has stricter requirements than most other kernel subsystems on
53 what the userspace side for new uAPI needs to look like. This section here
54 explains what exactly those requirements are, and why they exist.
56 The short summary is that any addition of DRM uAPI requires corresponding
57 open-sourced userspace patches, and those patches must be reviewed and ready for
58 merging into a suitable and canonical upstream project.
60 GFX devices (both display and render/GPU side) are really complex bits of
61 hardware, with userspace and kernel by necessity having to work together really
62 closely. The interfaces, for rendering and modesetting, must be extremely wide
63 and flexible, and therefore it is almost always impossible to precisely define
64 them for every possible corner case. This in turn makes it really practically
65 infeasible to differentiate between behaviour that's required by userspace, and
66 which must not be changed to avoid regressions, and behaviour which is only an
67 accidental artifact of the current implementation.
69 Without access to the full source code of all userspace users that means it
70 becomes impossible to change the implementation details, since userspace could
71 depend upon the accidental behaviour of the current implementation in minute
72 details. And debugging such regressions without access to source code is pretty
73 much impossible. As a consequence this means:
75 - The Linux kernel's "no regression" policy holds in practice only for
76 open-source userspace of the DRM subsystem. DRM developers are perfectly fine
77 if closed-source blob drivers in userspace use the same uAPI as the open
78 drivers, but they must do so in the exact same way as the open drivers.
79 Creative (ab)use of the interfaces will, and in the past routinely has, lead
82 - Any new userspace interface must have an open-source implementation as
83 demonstration vehicle.
85 The other reason for requiring open-source userspace is uAPI review. Since the
86 kernel and userspace parts of a GFX stack must work together so closely, code
87 review can only assess whether a new interface achieves its goals by looking at
88 both sides. Making sure that the interface indeed covers the use-case fully
89 leads to a few additional requirements:
91 - The open-source userspace must not be a toy/test application, but the real
92 thing. Specifically it needs to handle all the usual error and corner cases.
93 These are often the places where new uAPI falls apart and hence essential to
94 assess the fitness of a proposed interface.
96 - The userspace side must be fully reviewed and tested to the standards of that
97 userspace project. For e.g. mesa this means piglit testcases and review on the
98 mailing list. This is again to ensure that the new interface actually gets the
99 job done. The userspace-side reviewer should also provide an Acked-by on the
100 kernel uAPI patch indicating that they believe the proposed uAPI is sound and
101 sufficiently documented and validated for userspace's consumption.
103 - The userspace patches must be against the canonical upstream, not some vendor
104 fork. This is to make sure that no one cheats on the review and testing
105 requirements by doing a quick fork.
107 - The kernel patch can only be merged after all the above requirements are met,
108 but it **must** be merged to either drm-next or drm-misc-next **before** the
109 userspace patches land. uAPI always flows from the kernel, doing things the
110 other way round risks divergence of the uAPI definitions and header files.
112 These are fairly steep requirements, but have grown out from years of shared
113 pain and experience with uAPI added hastily, and almost always regretted about
114 just as fast. GFX devices change really fast, requiring a paradigm shift and
115 entire new set of uAPI interfaces every few years at least. Together with the
116 Linux kernel's guarantee to keep existing userspace running for 10+ years this
117 is already rather painful for the DRM subsystem, with multiple different uAPIs
118 for the same thing co-existing. If we add a few more complete mistakes into the
119 mix every year it would be entirely unmanageable.
126 DRM core provides multiple character-devices for user-space to use.
127 Depending on which device is opened, user-space can perform a different
128 set of operations (mainly ioctls). The primary node is always created
129 and called card<num>. Additionally, a currently unused control node,
130 called controlD<num> is also created. The primary node provides all
131 legacy operations and historically was the only interface used by
132 userspace. With KMS, the control node was introduced. However, the
133 planned KMS control interface has never been written and so the control
134 node stays unused to date.
136 With the increased use of offscreen renderers and GPGPU applications,
137 clients no longer require running compositors or graphics servers to
138 make use of a GPU. But the DRM API required unprivileged clients to
139 authenticate to a DRM-Master prior to getting GPU access. To avoid this
140 step and to grant clients GPU access without authenticating, render
141 nodes were introduced. Render nodes solely serve render clients, that
142 is, no modesetting or privileged ioctls can be issued on render nodes.
143 Only non-global rendering commands are allowed. If a driver supports
144 render nodes, it must advertise it via the DRIVER_RENDER DRM driver
145 capability. If not supported, the primary node must be used for render
146 clients together with the legacy drmAuth authentication procedure.
148 If a driver advertises render node support, DRM core will create a
149 separate render node called renderD<num>. There will be one render node
150 per device. No ioctls except PRIME-related ioctls will be allowed on
151 this node. Especially GEM_OPEN will be explicitly prohibited. For a
152 complete list of driver-independent ioctls that can be used on render
153 nodes, see the ioctls marked DRM_RENDER_ALLOW in drm_ioctl.c Render
154 nodes are designed to avoid the buffer-leaks, which occur if clients
155 guess the flink names or mmap offsets on the legacy interface.
156 Additionally to this basic interface, drivers must mark their
157 driver-dependent render-only ioctls as DRM_RENDER_ALLOW so render
158 clients can use them. Driver authors must be careful not to allow any
159 privileged ioctls on render nodes.
161 With render nodes, user-space can now control access to the render node
162 via basic file-system access-modes. A running graphics server which
163 authenticates clients on the privileged primary/legacy node is no longer
164 required. Instead, a client can open the render node and is immediately
165 granted GPU access. Communication between clients (or servers) is done
166 via PRIME. FLINK from render node to legacy node is not supported. New
167 clients must not use the insecure FLINK interface.
169 Besides dropping all modeset/global ioctls, render nodes also drop the
170 DRM-Master concept. There is no reason to associate render clients with
171 a DRM-Master as they are independent of any graphics server. Besides,
172 they must work without any running master, anyway. Drivers must be able
173 to run without a master object if they support render nodes. If, on the
174 other hand, a driver requires shared state between clients which is
175 visible to user-space and accessible beyond open-file boundaries, they
176 cannot support render nodes.
182 The following is the plan. Implementation is not there yet
185 Graphics devices (display and/or render) may be connected via USB (e.g.
186 display adapters or docking stations) or Thunderbolt (e.g. eGPU). An end
187 user is able to hot-unplug this kind of devices while they are being
188 used, and expects that the very least the machine does not crash. Any
189 damage from hot-unplugging a DRM device needs to be limited as much as
190 possible and userspace must be given the chance to handle it if it wants
191 to. Ideally, unplugging a DRM device still lets a desktop continue to
192 run, but that is going to need explicit support throughout the whole
193 graphics stack: from kernel and userspace drivers, through display
194 servers, via window system protocols, and in applications and libraries.
196 Other scenarios that should lead to the same are: unrecoverable GPU
197 crash, PCI device disappearing off the bus, or forced unbind of a driver
198 from the physical device.
200 In other words, from userspace perspective everything needs to keep on
201 working more or less, until userspace stops using the disappeared DRM
202 device and closes it completely. Userspace will learn of the device
203 disappearance from the device removed uevent, ioctls returning ENODEV
204 (or driver-specific ioctls returning driver-specific things), or open()
207 Only after userspace has closed all relevant DRM device and dmabuf file
208 descriptors and removed all mmaps, the DRM driver can tear down its
209 instance for the device that no longer exists. If the same physical
210 device somehow comes back in the mean time, it shall be a new DRM
213 Similar to PIDs, chardev minor numbers are not recycled immediately. A
214 new DRM device always picks the next free minor number compared to the
215 previous one allocated, and wraps around when minor numbers are
218 The goal raises at least the following requirements for the kernel and
221 Requirements for KMS UAPI
222 -------------------------
224 - KMS connectors must change their status to disconnected.
226 - Legacy modesets and pageflips, and atomic commits, both real and
227 TEST_ONLY, and any other ioctls either fail with ENODEV or fake
230 - Pending non-blocking KMS operations deliver the DRM events userspace
231 is expecting. This applies also to ioctls that faked success.
233 - open() on a device node whose underlying device has disappeared will
236 - Attempting to create a DRM lease on a disappeared DRM device will
237 fail with ENODEV. Existing DRM leases remain and work as listed
240 Requirements for Render and Cross-Device UAPI
241 ---------------------------------------------
243 - All GPU jobs that can no longer run must have their fences
244 force-signalled to avoid inflicting hangs on userspace.
245 The associated error code is ENODEV.
247 - Some userspace APIs already define what should happen when the device
248 disappears (OpenGL, GL ES: `GL_KHR_robustness`_; `Vulkan`_:
249 VK_ERROR_DEVICE_LOST; etc.). DRM drivers are free to implement this
250 behaviour the way they see best, e.g. returning failures in
251 driver-specific ioctls and handling those in userspace drivers, or
252 rely on uevents, and so on.
254 - dmabuf which point to memory that has disappeared will either fail to
255 import with ENODEV or continue to be successfully imported if it would
256 have succeeded before the disappearance. See also about memory maps
257 below for already imported dmabufs.
259 - Attempting to import a dmabuf to a disappeared device will either fail
260 with ENODEV or succeed if it would have succeeded without the
263 - open() on a device node whose underlying device has disappeared will
266 .. _GL_KHR_robustness: https://www.khronos.org/registry/OpenGL/extensions/KHR/KHR_robustness.txt
267 .. _Vulkan: https://www.khronos.org/vulkan/
269 Requirements for Memory Maps
270 ----------------------------
272 Memory maps have further requirements that apply to both existing maps
273 and maps created after the device has disappeared. If the underlying
274 memory disappears, the map is created or modified such that reads and
275 writes will still complete successfully but the result is undefined.
276 This applies to both userspace mmap()'d memory and memory pointed to by
277 dmabuf which might be mapped to other devices (cross-device dmabuf
280 Raising SIGBUS is not an option, because userspace cannot realistically
281 handle it. Signal handlers are global, which makes them extremely
282 difficult to use correctly from libraries like those that Mesa produces.
283 Signal handlers are not composable, you can't have different handlers
284 for GPU1 and GPU2 from different vendors, and a third handler for
285 mmapped regular files. Threads cause additional pain with signal
288 .. _drm_driver_ioctl:
290 IOCTL Support on Device Nodes
291 =============================
293 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
294 :doc: driver specific ioctls
296 Recommended IOCTL Return Values
297 -------------------------------
299 In theory a driver's IOCTL callback is only allowed to return very few error
300 codes. In practice it's good to abuse a few more. This section documents common
301 practice within the DRM subsystem:
304 Strictly this should only be used when a file doesn't exist e.g. when
305 calling the open() syscall. We reuse that to signal any kind of object
306 lookup failure, e.g. for unknown GEM buffer object handles, unknown KMS
307 object handles and similar cases.
310 Some drivers use this to differentiate "out of kernel memory" from "out
311 of VRAM". Sometimes also applies to other limited gpu resources used for
312 rendering (e.g. when you have a special limited compression buffer).
313 Sometimes resource allocation/reservation issues in command submission
314 IOCTLs are also signalled through EDEADLK.
316 Simply running out of kernel/system memory is signalled through ENOMEM.
319 Returned for an operation that is valid, but needs more privileges.
320 E.g. root-only or much more common, DRM master-only operations return
321 this when called by unpriviledged clients. There's no clear
322 difference between EACCES and EPERM.
325 The device is not present anymore or is not yet fully initialized.
328 Feature (like PRIME, modesetting, GEM) is not supported by the driver.
331 Remote failure, either a hardware transaction (like i2c), but also used
332 when the exporting driver of a shared dma-buf or fence doesn't support a
336 DRM drivers assume that userspace restarts all IOCTLs. Any DRM IOCTL can
337 return EINTR and in such a case should be restarted with the IOCTL
338 parameters left unchanged.
341 The GPU died and couldn't be resurrected through a reset. Modesetting
342 hardware failures are signalled through the "link status" connector
346 Catch-all for anything that is an invalid argument combination which
349 IOCTL also use other error codes like ETIME, EFAULT, EBUSY, ENOTTY but their
350 usage is in line with the common meanings. The above list tries to just document
351 DRM specific patterns. Note that ENOTTY has the slightly unintuitive meaning of
352 "this IOCTL does not exist", and is used exactly as such in DRM.
354 .. kernel-doc:: include/drm/drm_ioctl.h
357 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
360 .. kernel-doc:: drivers/gpu/drm/drm_ioc32.c
363 Testing and validation
364 ======================
366 Testing Requirements for userspace API
367 --------------------------------------
369 New cross-driver userspace interface extensions, like new IOCTL, new KMS
370 properties, new files in sysfs or anything else that constitutes an API change
371 should have driver-agnostic testcases in IGT for that feature, if such a test
372 can be reasonably made using IGT for the target hardware.
374 Validating changes with IGT
375 ---------------------------
377 There's a collection of tests that aims to cover the whole functionality of
378 DRM drivers and that can be used to check that changes to DRM drivers or the
379 core don't regress existing functionality. This test suite is called IGT and
380 its code and instructions to build and run can be found in
381 https://gitlab.freedesktop.org/drm/igt-gpu-tools/.
383 Using VKMS to test DRM API
384 --------------------------
386 VKMS is a software-only model of a KMS driver that is useful for testing
387 and for running compositors. VKMS aims to enable a virtual display without
388 the need for a hardware display capability. These characteristics made VKMS
389 a perfect tool for validating the DRM core behavior and also support the
390 compositor developer. VKMS makes it possible to test DRM functions in a
391 virtual machine without display, simplifying the validation of some of the
394 To Validate changes in DRM API with VKMS, start setting the kernel: make
395 sure to enable VKMS module; compile the kernel with the VKMS enabled and
396 install it in the target machine. VKMS can be run in a Virtual Machine
397 (QEMU, virtme or similar). It's recommended the use of KVM with the minimum
398 of 1GB of RAM and four cores.
400 It's possible to run the IGT-tests in a VM in two ways:
402 1. Use IGT inside a VM
403 2. Use IGT from the host machine and write the results in a shared directory.
405 Following is an example of using a VM with a shared directory with
406 the host machine to run igt-tests. This example uses virtme::
408 $ virtme-run --rwdir /path/for/shared_dir --kdir=path/for/kernel/directory --mods=auto
410 Run the igt-tests in the guest machine. This example runs the 'kms_flip'
413 $ /path/for/igt-gpu-tools/scripts/run-tests.sh -p -s -t "kms_flip.*" -v
415 In this example, instead of building the igt_runner, Piglit is used
416 (-p option). It creates an HTML summary of the test results and saves
417 them in the folder "igt-gpu-tools/results". It executes only the igt-tests
418 matching the -t option.
423 .. kernel-doc:: drivers/gpu/drm/drm_debugfs_crc.c
426 .. kernel-doc:: drivers/gpu/drm/drm_debugfs_crc.c
432 .. kernel-doc:: include/drm/drm_debugfs.h
435 .. kernel-doc:: drivers/gpu/drm/drm_debugfs.c
441 .. kernel-doc:: drivers/gpu/drm/drm_sysfs.c
444 .. kernel-doc:: drivers/gpu/drm/drm_sysfs.c
448 VBlank event handling
449 =====================
451 The DRM core exposes two vertical blank related ioctls:
453 DRM_IOCTL_WAIT_VBLANK
454 This takes a struct drm_wait_vblank structure as its argument, and
455 it is used to block or request a signal when a specified vblank
458 DRM_IOCTL_MODESET_CTL
459 This was only used for user-mode-settind drivers around modesetting
460 changes to allow the kernel to update the vblank interrupt after
461 mode setting, since on many devices the vertical blank counter is
462 reset to 0 at some point during modeset. Modern drivers should not
463 call this any more since with kernel mode setting it is a no-op.
465 Userspace API Structures
466 ========================
468 .. kernel-doc:: include/uapi/drm/drm_mode.h
476 CRTC's have both an object ID and an index, and they are not the same thing.
477 The index is used in cases where a densely packed identifier for a CRTC is
478 needed, for instance a bitmask of CRTC's. The member possible_crtcs of struct
479 drm_mode_get_plane is an example.
481 DRM_IOCTL_MODE_GETRESOURCES populates a structure with an array of CRTC ID's,
482 and the CRTC index is its position in this array.
484 .. kernel-doc:: include/uapi/drm/drm.h
487 .. kernel-doc:: include/uapi/drm/drm_mode.h