5 The DRM core exports several interfaces to applications, generally
6 intended to be used through corresponding libdrm wrapper functions. In
7 addition, drivers export device-specific interfaces for use by userspace
8 drivers & device-aware applications through ioctls and sysfs files.
10 External interfaces include: memory mapping, context management, DMA
11 operations, AGP management, vblank control, fence management, memory
12 management, and output management.
14 Cover generic ioctls and sysfs layout here. We only need high-level
15 info, since man pages should cover the rest.
20 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
21 :doc: getunique and setversion story
26 Primary Nodes, DRM Master and Authentication
27 ============================================
29 .. kernel-doc:: drivers/gpu/drm/drm_auth.c
30 :doc: master and authentication
32 .. kernel-doc:: drivers/gpu/drm/drm_auth.c
35 .. kernel-doc:: include/drm/drm_auth.h
38 Open-Source Userspace Requirements
39 ==================================
41 The DRM subsystem has stricter requirements than most other kernel subsystems on
42 what the userspace side for new uAPI needs to look like. This section here
43 explains what exactly those requirements are, and why they exist.
45 The short summary is that any addition of DRM uAPI requires corresponding
46 open-sourced userspace patches, and those patches must be reviewed and ready for
47 merging into a suitable and canonical upstream project.
49 GFX devices (both display and render/GPU side) are really complex bits of
50 hardware, with userspace and kernel by necessity having to work together really
51 closely. The interfaces, for rendering and modesetting, must be extremely wide
52 and flexible, and therefore it is almost always impossible to precisely define
53 them for every possible corner case. This in turn makes it really practically
54 infeasible to differentiate between behaviour that's required by userspace, and
55 which must not be changed to avoid regressions, and behaviour which is only an
56 accidental artifact of the current implementation.
58 Without access to the full source code of all userspace users that means it
59 becomes impossible to change the implementation details, since userspace could
60 depend upon the accidental behaviour of the current implementation in minute
61 details. And debugging such regressions without access to source code is pretty
62 much impossible. As a consequence this means:
64 - The Linux kernel's "no regression" policy holds in practice only for
65 open-source userspace of the DRM subsystem. DRM developers are perfectly fine
66 if closed-source blob drivers in userspace use the same uAPI as the open
67 drivers, but they must do so in the exact same way as the open drivers.
68 Creative (ab)use of the interfaces will, and in the past routinely has, lead
71 - Any new userspace interface must have an open-source implementation as
72 demonstration vehicle.
74 The other reason for requiring open-source userspace is uAPI review. Since the
75 kernel and userspace parts of a GFX stack must work together so closely, code
76 review can only assess whether a new interface achieves its goals by looking at
77 both sides. Making sure that the interface indeed covers the use-case fully
78 leads to a few additional requirements:
80 - The open-source userspace must not be a toy/test application, but the real
81 thing. Specifically it needs to handle all the usual error and corner cases.
82 These are often the places where new uAPI falls apart and hence essential to
83 assess the fitness of a proposed interface.
85 - The userspace side must be fully reviewed and tested to the standards of that
86 userspace project. For e.g. mesa this means piglit testcases and review on the
87 mailing list. This is again to ensure that the new interface actually gets the
90 - The userspace patches must be against the canonical upstream, not some vendor
91 fork. This is to make sure that no one cheats on the review and testing
92 requirements by doing a quick fork.
94 - The kernel patch can only be merged after all the above requirements are met,
95 but it **must** be merged **before** the userspace patches land. uAPI always flows
96 from the kernel, doing things the other way round risks divergence of the uAPI
97 definitions and header files.
99 These are fairly steep requirements, but have grown out from years of shared
100 pain and experience with uAPI added hastily, and almost always regretted about
101 just as fast. GFX devices change really fast, requiring a paradigm shift and
102 entire new set of uAPI interfaces every few years at least. Together with the
103 Linux kernel's guarantee to keep existing userspace running for 10+ years this
104 is already rather painful for the DRM subsystem, with multiple different uAPIs
105 for the same thing co-existing. If we add a few more complete mistakes into the
106 mix every year it would be entirely unmanageable.
113 DRM core provides multiple character-devices for user-space to use.
114 Depending on which device is opened, user-space can perform a different
115 set of operations (mainly ioctls). The primary node is always created
116 and called card<num>. Additionally, a currently unused control node,
117 called controlD<num> is also created. The primary node provides all
118 legacy operations and historically was the only interface used by
119 userspace. With KMS, the control node was introduced. However, the
120 planned KMS control interface has never been written and so the control
121 node stays unused to date.
123 With the increased use of offscreen renderers and GPGPU applications,
124 clients no longer require running compositors or graphics servers to
125 make use of a GPU. But the DRM API required unprivileged clients to
126 authenticate to a DRM-Master prior to getting GPU access. To avoid this
127 step and to grant clients GPU access without authenticating, render
128 nodes were introduced. Render nodes solely serve render clients, that
129 is, no modesetting or privileged ioctls can be issued on render nodes.
130 Only non-global rendering commands are allowed. If a driver supports
131 render nodes, it must advertise it via the DRIVER_RENDER DRM driver
132 capability. If not supported, the primary node must be used for render
133 clients together with the legacy drmAuth authentication procedure.
135 If a driver advertises render node support, DRM core will create a
136 separate render node called renderD<num>. There will be one render node
137 per device. No ioctls except PRIME-related ioctls will be allowed on
138 this node. Especially GEM_OPEN will be explicitly prohibited. Render
139 nodes are designed to avoid the buffer-leaks, which occur if clients
140 guess the flink names or mmap offsets on the legacy interface.
141 Additionally to this basic interface, drivers must mark their
142 driver-dependent render-only ioctls as DRM_RENDER_ALLOW so render
143 clients can use them. Driver authors must be careful not to allow any
144 privileged ioctls on render nodes.
146 With render nodes, user-space can now control access to the render node
147 via basic file-system access-modes. A running graphics server which
148 authenticates clients on the privileged primary/legacy node is no longer
149 required. Instead, a client can open the render node and is immediately
150 granted GPU access. Communication between clients (or servers) is done
151 via PRIME. FLINK from render node to legacy node is not supported. New
152 clients must not use the insecure FLINK interface.
154 Besides dropping all modeset/global ioctls, render nodes also drop the
155 DRM-Master concept. There is no reason to associate render clients with
156 a DRM-Master as they are independent of any graphics server. Besides,
157 they must work without any running master, anyway. Drivers must be able
158 to run without a master object if they support render nodes. If, on the
159 other hand, a driver requires shared state between clients which is
160 visible to user-space and accessible beyond open-file boundaries, they
161 cannot support render nodes.
163 .. _drm_driver_ioctl:
165 IOCTL Support on Device Nodes
166 =============================
168 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
169 :doc: driver specific ioctls
171 Recommended IOCTL Return Values
172 -------------------------------
174 In theory a driver's IOCTL callback is only allowed to return very few error
175 codes. In practice it's good to abuse a few more. This section documents common
176 practice within the DRM subsystem:
179 Strictly this should only be used when a file doesn't exist e.g. when
180 calling the open() syscall. We reuse that to signal any kind of object
181 lookup failure, e.g. for unknown GEM buffer object handles, unknown KMS
182 object handles and similar cases.
185 Some drivers use this to differentiate "out of kernel memory" from "out
186 of VRAM". Sometimes also applies to other limited gpu resources used for
187 rendering (e.g. when you have a special limited compression buffer).
188 Sometimes resource allocation/reservation issues in command submission
189 IOCTLs are also signalled through EDEADLK.
191 Simply running out of kernel/system memory is signalled through ENOMEM.
194 Returned for an operation that is valid, but needs more privileges.
195 E.g. root-only or much more common, DRM master-only operations return
196 this when when called by unpriviledged clients. There's no clear
197 difference between EACCES and EPERM.
200 Feature (like PRIME, modesetting, GEM) is not supported by the driver.
203 Remote failure, either a hardware transaction (like i2c), but also used
204 when the exporting driver of a shared dma-buf or fence doesn't support a
208 DRM drivers assume that userspace restarts all IOCTLs. Any DRM IOCTL can
209 return EINTR and in such a case should be restarted with the IOCTL
210 parameters left unchanged.
213 The GPU died and couldn't be resurrected through a reset. Modesetting
214 hardware failures are signalled through the "link status" connector
218 Catch-all for anything that is an invalid argument combination which
221 IOCTL also use other error codes like ETIME, EFAULT, EBUSY, ENOTTY but their
222 usage is in line with the common meanings. The above list tries to just document
223 DRM specific patterns. Note that ENOTTY has the slightly unintuitive meaning of
224 "this IOCTL does not exist", and is used exactly as such in DRM.
226 .. kernel-doc:: include/drm/drm_ioctl.h
229 .. kernel-doc:: drivers/gpu/drm/drm_ioctl.c
232 .. kernel-doc:: drivers/gpu/drm/drm_ioc32.c
235 Testing and validation
236 ======================
238 Validating changes with IGT
239 ---------------------------
241 There's a collection of tests that aims to cover the whole functionality of
242 DRM drivers and that can be used to check that changes to DRM drivers or the
243 core don't regress existing functionality. This test suite is called IGT and
244 its code can be found in https://cgit.freedesktop.org/drm/igt-gpu-tools/.
246 To build IGT, start by installing its build dependencies. In Debian-based
249 # apt-get build-dep intel-gpu-tools
251 And in Fedora-based systems::
253 # dnf builddep intel-gpu-tools
255 Then clone the repository::
257 $ git clone git://anongit.freedesktop.org/drm/igt-gpu-tools
259 Configure the build system and start the build::
261 $ cd igt-gpu-tools && ./autogen.sh && make -j6
263 Download the piglit dependency::
265 $ ./scripts/run-tests.sh -d
269 $ ./scripts/run-tests.sh -t kms -t core -s
271 run-tests.sh is a wrapper around piglit that will execute the tests matching
272 the -t options. A report in HTML format will be available in
273 ./results/html/index.html. Results can be compared with piglit.
278 .. kernel-doc:: drivers/gpu/drm/drm_debugfs_crc.c
281 .. kernel-doc:: drivers/gpu/drm/drm_debugfs_crc.c
287 .. kernel-doc:: include/drm/drm_debugfs.h
290 .. kernel-doc:: drivers/gpu/drm/drm_debugfs.c
296 .. kernel-doc:: drivers/gpu/drm/drm_sysfs.c
299 .. kernel-doc:: drivers/gpu/drm/drm_sysfs.c
303 VBlank event handling
304 =====================
306 The DRM core exposes two vertical blank related ioctls:
308 DRM_IOCTL_WAIT_VBLANK
309 This takes a struct drm_wait_vblank structure as its argument, and
310 it is used to block or request a signal when a specified vblank
313 DRM_IOCTL_MODESET_CTL
314 This was only used for user-mode-settind drivers around modesetting
315 changes to allow the kernel to update the vblank interrupt after
316 mode setting, since on many devices the vertical blank counter is
317 reset to 0 at some point during modeset. Modern drivers should not
318 call this any more since with kernel mode setting it is a no-op.