xref: /linux/Documentation/gpu/todo.rst (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1.. _todo:
2
3=========
4TODO list
5=========
6
7This section contains a list of smaller janitorial tasks in the kernel DRM
8graphics subsystem useful as newbie projects. Or for slow rainy days.
9
10Subsystem-wide refactorings
11===========================
12
13De-midlayer drivers
14-------------------
15
16With the recent ``drm_bus`` cleanup patches for 3.17 it is no longer required
17to have a ``drm_bus`` structure set up. Drivers can directly set up the
18``drm_device`` structure instead of relying on bus methods in ``drm_usb.c``
19and ``drm_pci.c``. The goal is to get rid of the driver's ``->load`` /
20``->unload`` callbacks and open-code the load/unload sequence properly, using
21the new two-stage ``drm_device`` setup/teardown.
22
23Once all existing drivers are converted we can also remove those bus support
24files for USB and platform devices.
25
26All you need is a GPU for a non-converted driver (currently almost all of
27them, but also all the virtual ones used by KVM, so everyone qualifies).
28
29Contact: Daniel Vetter, Thierry Reding, respective driver maintainers
30
31Switch from reference/unreference to get/put
32--------------------------------------------
33
34For some reason DRM core uses ``reference``/``unreference`` suffixes for
35refcounting functions, but kernel uses ``get``/``put`` (e.g.
36``kref_get``/``put()``). It would be good to switch over for consistency, and
37it's shorter. Needs to be done in 3 steps for each pair of functions:
38
39* Create new ``get``/``put`` functions, define the old names as compatibility
40  wrappers
41* Switch over each file/driver using a cocci-generated spatch.
42* Once all users of the old names are gone, remove them.
43
44This way drivers/patches in the progress of getting merged won't break.
45
46Contact: Daniel Vetter
47
48Convert existing KMS drivers to atomic modesetting
49--------------------------------------------------
50
513.19 has the atomic modeset interfaces and helpers, so drivers can now be
52converted over. Modern compositors like Wayland or Surfaceflinger on Android
53really want an atomic modeset interface, so this is all about the bright
54future.
55
56There is a conversion guide for atomic and all you need is a GPU for a
57non-converted driver (again virtual HW drivers for KVM are still all
58suitable).
59
60As part of this drivers also need to convert to universal plane (which means
61exposing primary & cursor as proper plane objects). But that's much easier to
62do by directly using the new atomic helper driver callbacks.
63
64Contact: Daniel Vetter, respective driver maintainers
65
66Clean up the clipped coordination confusion around planes
67---------------------------------------------------------
68
69We have a helper to get this right with drm_plane_helper_check_update(), but
70it's not consistently used. This should be fixed, preferrably in the atomic
71helpers (and drivers then moved over to clipped coordinates). Probably the
72helper should also be moved from drm_plane_helper.c to the atomic helpers, to
73avoid confusion - the other helpers in that file are all deprecated legacy
74helpers.
75
76Contact: Ville Syrjälä, Daniel Vetter, driver maintainers
77
78Convert early atomic drivers to async commit helpers
79----------------------------------------------------
80
81For the first year the atomic modeset helpers didn't support asynchronous /
82nonblocking commits, and every driver had to hand-roll them. This is fixed
83now, but there's still a pile of existing drivers that easily could be
84converted over to the new infrastructure.
85
86One issue with the helpers is that they require that drivers handle completion
87events for atomic commits correctly. But fixing these bugs is good anyway.
88
89Contact: Daniel Vetter, respective driver maintainers
90
91Better manual-upload support for atomic
92---------------------------------------
93
94This would be especially useful for tinydrm:
95
96- Add a struct drm_rect dirty_clip to drm_crtc_state. When duplicating the
97  crtc state, clear that to the max values, x/y = 0 and w/h = MAX_INT, in
98  __drm_atomic_helper_crtc_duplicate_state().
99
100- Move tinydrm_merge_clips into drm_framebuffer.c, dropping the tinydrm\_
101  prefix ofc and using drm_fb\_. drm_framebuffer.c makes sense since this
102  is a function useful to implement the fb->dirty function.
103
104- Create a new drm_fb_dirty function which does essentially what e.g.
105  mipi_dbi_fb_dirty does. You can use e.g. drm_atomic_helper_update_plane as the
106  template. But instead of doing a simple full-screen plane update, this new
107  helper also sets crtc_state->dirty_clip to the right coordinates. And of
108  course it needs to check whether the fb is actually active (and maybe where),
109  so there's some book-keeping involved. There's also some good fun involved in
110  scaling things appropriately. For that case we might simply give up and
111  declare the entire area covered by the plane as dirty.
112
113Contact: Noralf Trønnes, Daniel Vetter
114
115Fallout from atomic KMS
116-----------------------
117
118``drm_atomic_helper.c`` provides a batch of functions which implement legacy
119IOCTLs on top of the new atomic driver interface. Which is really nice for
120gradual conversion of drivers, but unfortunately the semantic mismatches are
121a bit too severe. So there's some follow-up work to adjust the function
122interfaces to fix these issues:
123
124* atomic needs the lock acquire context. At the moment that's passed around
125  implicitly with some horrible hacks, and it's also allocate with
126  ``GFP_NOFAIL`` behind the scenes. All legacy paths need to start allocating
127  the acquire context explicitly on stack and then also pass it down into
128  drivers explicitly so that the legacy-on-atomic functions can use them.
129
130  Except for some driver code this is done.
131
132* A bunch of the vtable hooks are now in the wrong place: DRM has a split
133  between core vfunc tables (named ``drm_foo_funcs``), which are used to
134  implement the userspace ABI. And then there's the optional hooks for the
135  helper libraries (name ``drm_foo_helper_funcs``), which are purely for
136  internal use. Some of these hooks should be move from ``_funcs`` to
137  ``_helper_funcs`` since they are not part of the core ABI. There's a
138  ``FIXME`` comment in the kerneldoc for each such case in ``drm_crtc.h``.
139
140* There's a new helper ``drm_atomic_helper_best_encoder()`` which could be
141  used by all atomic drivers which don't select the encoder for a given
142  connector at runtime. That's almost all of them, and would allow us to get
143  rid of a lot of ``best_encoder`` boilerplate in drivers.
144
145  This was almost done, but new drivers added a few more cases again.
146
147Contact: Daniel Vetter
148
149Get rid of dev->struct_mutex from GEM drivers
150---------------------------------------------
151
152``dev->struct_mutex`` is the Big DRM Lock from legacy days and infested
153everything. Nowadays in modern drivers the only bit where it's mandatory is
154serializing GEM buffer object destruction. Which unfortunately means drivers
155have to keep track of that lock and either call ``unreference`` or
156``unreference_locked`` depending upon context.
157
158Core GEM doesn't have a need for ``struct_mutex`` any more since kernel 4.8,
159and there's a ``gem_free_object_unlocked`` callback for any drivers which are
160entirely ``struct_mutex`` free.
161
162For drivers that need ``struct_mutex`` it should be replaced with a driver-
163private lock. The tricky part is the BO free functions, since those can't
164reliably take that lock any more. Instead state needs to be protected with
165suitable subordinate locks or some cleanup work pushed to a worker thread. For
166performance-critical drivers it might also be better to go with a more
167fine-grained per-buffer object and per-context lockings scheme. Currently the
168following drivers still use ``struct_mutex``: ``msm``, ``omapdrm`` and
169``udl``.
170
171Contact: Daniel Vetter, respective driver maintainers
172
173Convert instances of dev_info/dev_err/dev_warn to their DRM_DEV_* equivalent
174----------------------------------------------------------------------------
175
176For drivers which could have multiple instances, it is necessary to
177differentiate between which is which in the logs. Since DRM_INFO/WARN/ERROR
178don't do this, drivers used dev_info/warn/err to make this differentiation. We
179now have DRM_DEV_* variants of the drm print macros, so we can start to convert
180those drivers back to using drm-formwatted specific log messages.
181
182Before you start this conversion please contact the relevant maintainers to make
183sure your work will be merged - not everyone agrees that the DRM dmesg macros
184are better.
185
186Contact: Sean Paul, Maintainer of the driver you plan to convert
187
188Convert drivers to use simple modeset suspend/resume
189----------------------------------------------------
190
191Most drivers (except i915 and nouveau) that use
192drm_atomic_helper_suspend/resume() can probably be converted to use
193drm_mode_config_helper_suspend/resume().
194
195Contact: Maintainer of the driver you plan to convert
196
197Convert drivers to use drm_fb_helper_fbdev_setup/teardown()
198-----------------------------------------------------------
199
200Most drivers can use drm_fb_helper_fbdev_setup() except maybe:
201
202- amdgpu which has special logic to decide whether to call
203  drm_helper_disable_unused_functions()
204
205- armada which isn't atomic and doesn't call
206  drm_helper_disable_unused_functions()
207
208- i915 which calls drm_fb_helper_initial_config() in a worker
209
210Drivers that use drm_framebuffer_remove() to clean up the fbdev framebuffer can
211probably use drm_fb_helper_fbdev_teardown().
212
213Contact: Maintainer of the driver you plan to convert
214
215Clean up mmap forwarding
216------------------------
217
218A lot of drivers forward gem mmap calls to dma-buf mmap for imported buffers.
219And also a lot of them forward dma-buf mmap to the gem mmap implementations.
220Would be great to refactor this all into a set of small common helpers.
221
222Contact: Daniel Vetter
223
224Put a reservation_object into drm_gem_object
225--------------------------------------------
226
227This would remove the need for the ->gem_prime_res_obj callback. It would also
228allow us to implement generic helpers for waiting for a bo, allowing for quite a
229bit of refactoring in the various wait ioctl implementations.
230
231Contact: Daniel Vetter
232
233idr_init_base()
234---------------
235
236DRM core&drivers uses a lot of idr (integer lookup directories) for mapping
237userspace IDs to internal objects, and in most places ID=0 means NULL and hence
238is never used. Switching to idr_init_base() for these would make the idr more
239efficient.
240
241Contact: Daniel Vetter
242
243Core refactorings
244=================
245
246Clean up the DRM header mess
247----------------------------
248
249Currently the DRM subsystem has only one global header, ``drmP.h``. This is
250used both for functions exported to helper libraries and drivers and functions
251only used internally in the ``drm.ko`` module. The goal would be to move all
252header declarations not needed outside of ``drm.ko`` into
253``drivers/gpu/drm/drm_*_internal.h`` header files. ``EXPORT_SYMBOL`` also
254needs to be dropped for these functions.
255
256This would nicely tie in with the below task to create kerneldoc after the API
257is cleaned up. Or with the "hide legacy cruft better" task.
258
259Note that this is well in progress, but ``drmP.h`` is still huge. The updated
260plan is to switch to per-file driver API headers, which will also structure
261the kerneldoc better. This should also allow more fine-grained ``#include``
262directives.
263
264In the end no .c file should need to include ``drmP.h`` anymore.
265
266Contact: Daniel Vetter
267
268Add missing kerneldoc for exported functions
269--------------------------------------------
270
271The DRM reference documentation is still lacking kerneldoc in a few areas. The
272task would be to clean up interfaces like moving functions around between
273files to better group them and improving the interfaces like dropping return
274values for functions that never fail. Then write kerneldoc for all exported
275functions and an overview section and integrate it all into the drm book.
276
277See https://dri.freedesktop.org/docs/drm/ for what's there already.
278
279Contact: Daniel Vetter
280
281Hide legacy cruft better
282------------------------
283
284Way back DRM supported only drivers which shadow-attached to PCI devices with
285userspace or fbdev drivers setting up outputs. Modern DRM drivers take charge
286of the entire device, you can spot them with the DRIVER_MODESET flag.
287
288Unfortunately there's still large piles of legacy code around which needs to
289be hidden so that driver writers don't accidentally end up using it. And to
290prevent security issues in those legacy IOCTLs from being exploited on modern
291drivers. This has multiple possible subtasks:
292
293* Extract support code for legacy features into a ``drm-legacy.ko`` kernel
294  module and compile it only when one of the legacy drivers is enabled.
295
296This is mostly done, the only thing left is to split up ``drm_irq.c`` into
297legacy cruft and the parts needed by modern KMS drivers.
298
299Contact: Daniel Vetter
300
301Make panic handling work
302------------------------
303
304This is a really varied tasks with lots of little bits and pieces:
305
306* The panic path can't be tested currently, leading to constant breaking. The
307  main issue here is that panics can be triggered from hardirq contexts and
308  hence all panic related callback can run in hardirq context. It would be
309  awesome if we could test at least the fbdev helper code and driver code by
310  e.g. trigger calls through drm debugfs files. hardirq context could be
311  achieved by using an IPI to the local processor.
312
313* There's a massive confusion of different panic handlers. DRM fbdev emulation
314  helpers have one, but on top of that the fbcon code itself also has one. We
315  need to make sure that they stop fighting over each another.
316
317* ``drm_can_sleep()`` is a mess. It hides real bugs in normal operations and
318  isn't a full solution for panic paths. We need to make sure that it only
319  returns true if there's a panic going on for real, and fix up all the
320  fallout.
321
322* The panic handler must never sleep, which also means it can't ever
323  ``mutex_lock()``. Also it can't grab any other lock unconditionally, not
324  even spinlocks (because NMI and hardirq can panic too). We need to either
325  make sure to not call such paths, or trylock everything. Really tricky.
326
327* For the above locking troubles reasons it's pretty much impossible to
328  attempt a synchronous modeset from panic handlers. The only thing we could
329  try to achive is an atomic ``set_base`` of the primary plane, and hope that
330  it shows up. Everything else probably needs to be delayed to some worker or
331  something else which happens later on. Otherwise it just kills the box
332  harder, prevent the panic from going out on e.g. netconsole.
333
334* There's also proposal for a simplied DRM console instead of the full-blown
335  fbcon and DRM fbdev emulation. Any kind of panic handling tricks should
336  obviously work for both console, in case we ever get kmslog merged.
337
338Contact: Daniel Vetter
339
340Clean up the debugfs support
341----------------------------
342
343There's a bunch of issues with it:
344
345- The drm_info_list ->show() function doesn't even bother to cast to the drm
346  structure for you. This is lazy.
347
348- We probably want to have some support for debugfs files on crtc/connectors and
349  maybe other kms objects directly in core. There's even drm_print support in
350  the funcs for these objects to dump kms state, so it's all there. And then the
351  ->show() functions should obviously give you a pointer to the right object.
352
353- The drm_info_list stuff is centered on drm_minor instead of drm_device. For
354  anything we want to print drm_device (or maybe drm_file) is the right thing.
355
356- The drm_driver->debugfs_init hooks we have is just an artifact of the old
357  midlayered load sequence. DRM debugfs should work more like sysfs, where you
358  can create properties/files for an object anytime you want, and the core
359  takes care of publishing/unpuplishing all the files at register/unregister
360  time. Drivers shouldn't need to worry about these technicalities, and fixing
361  this (together with the drm_minor->drm_device move) would allow us to remove
362  debugfs_init.
363
364Contact: Daniel Vetter
365
366KMS cleanups
367------------
368
369Some of these date from the very introduction of KMS in 2008 ...
370
371- drm_mode_config.crtc_idr is misnamed, since it contains all KMS object. Should
372  be renamed to drm_mode_config.object_idr.
373
374- drm_display_mode doesn't need to be derived from drm_mode_object. That's
375  leftovers from older (never merged into upstream) KMS designs where modes
376  where set using their ID, including support to add/remove modes.
377
378Better Testing
379==============
380
381Enable trinity for DRM
382----------------------
383
384And fix up the fallout. Should be really interesting ...
385
386Make KMS tests in i-g-t generic
387-------------------------------
388
389The i915 driver team maintains an extensive testsuite for the i915 DRM driver,
390including tons of testcases for corner-cases in the modesetting API. It would
391be awesome if those tests (at least the ones not relying on Intel-specific GEM
392features) could be made to run on any KMS driver.
393
394Basic work to run i-g-t tests on non-i915 is done, what's now missing is mass-
395converting things over. For modeset tests we also first need a bit of
396infrastructure to use dumb buffers for untiled buffers, to be able to run all
397the non-i915 specific modeset tests.
398
399Contact: Daniel Vetter
400
401Create a virtual KMS driver for testing (vkms)
402----------------------------------------------
403
404With all the latest helpers it should be fairly simple to create a virtual KMS
405driver useful for testing, or for running X or similar on headless machines
406(to be able to still use the GPU). This would be similar to vgem, but aimed at
407the modeset side.
408
409Once the basics are there there's tons of possibilities to extend it.
410
411Contact: Daniel Vetter
412
413Driver Specific
414===============
415
416tinydrm
417-------
418
419Tinydrm is the helper driver for really simple fb drivers. The goal is to make
420those drivers as simple as possible, so lots of room for refactoring:
421
422- backlight helpers, probably best to put them into a new drm_backlight.c.
423  This is because drivers/video is de-facto unmaintained. We could also
424  move drivers/video/backlight to drivers/gpu/backlight and take it all
425  over within drm-misc, but that's more work. Backlight helpers require a fair
426  bit of reworking and refactoring. A simple example is the enabling of a backlight.
427  Tinydrm has helpers for this. It would be good if other drivers can also use the
428  helper. However, there are various cases we need to consider i.e different
429  drivers seem to have different ways of enabling/disabling a backlight.
430  We also need to consider the backlight drivers (like gpio_backlight). The situation
431  is further complicated by the fact that the backlight is tied to fbdev
432  via fb_notifier_callback() which has complicated logic. For further details, refer
433  to the following discussion thread:
434  https://groups.google.com/forum/#!topic/outreachy-kernel/8rBe30lwtdA
435
436- spi helpers, probably best put into spi core/helper code. Thierry said
437  the spi maintainer is fast&reactive, so shouldn't be a big issue.
438
439- extract the mipi-dbi helper (well, the non-tinydrm specific parts at
440  least) into a separate helper, like we have for mipi-dsi already. Or follow
441  one of the ideas for having a shared dsi/dbi helper, abstracting away the
442  transport details more.
443
444- tinydrm_gem_cma_prime_import_sg_table should probably go into the cma
445  helpers, as a _vmapped variant (since not every driver needs the vmap).
446  And tinydrm_gem_cma_free_object could the be merged into
447  drm_gem_cma_free_object().
448
449- tinydrm_fb_create we could move into drm_simple_pipe, only need to add
450  the fb_create hook to drm_simple_pipe_funcs, which would again simplify a
451  bunch of things (since it gives you a one-stop vfunc for simple drivers).
452
453- Quick aside: The unregister devm stuff is kinda getting the lifetimes of
454  a drm_device wrong. Doesn't matter, since everyone else gets it wrong
455  too :-)
456
457- also rework the drm_framebuffer_funcs->dirty hook wire-up, see above.
458
459Contact: Noralf Trønnes, Daniel Vetter
460
461AMD DC Display Driver
462---------------------
463
464AMD DC is the display driver for AMD devices starting with Vega. There has been
465a bunch of progress cleaning it up but there's still plenty of work to be done.
466
467See drivers/gpu/drm/amd/display/TODO for tasks.
468
469Contact: Harry Wentland, Alex Deucher
470
471i915
472----
473
474- Our early/late pm callbacks could be removed in favour of using
475  device_link_add to model the dependency between i915 and snd_had. See
476  https://dri.freedesktop.org/docs/drm/driver-api/device_link.html
477
478Outside DRM
479===========
480