xref: /openbmc/linux/Documentation/gpu/todo.rst (revision 2208f39c)
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
10Difficulty
11----------
12
13To make it easier task are categorized into different levels:
14
15Starter: Good tasks to get started with the DRM subsystem.
16
17Intermediate: Tasks which need some experience with working in the DRM
18subsystem, or some specific GPU/display graphics knowledge. For debugging issue
19it's good to have the relevant hardware (or a virtual driver set up) available
20for testing.
21
22Advanced: Tricky tasks that need fairly good understanding of the DRM subsystem
23and graphics topics. Generally need the relevant hardware for development and
24testing.
25
26Subsystem-wide refactorings
27===========================
28
29Remove custom dumb_map_offset implementations
30---------------------------------------------
31
32All GEM based drivers should be using drm_gem_create_mmap_offset() instead.
33Audit each individual driver, make sure it'll work with the generic
34implementation (there's lots of outdated locking leftovers in various
35implementations), and then remove it.
36
37Contact: Daniel Vetter, respective driver maintainers
38
39Level: Intermediate
40
41Convert existing KMS drivers to atomic modesetting
42--------------------------------------------------
43
443.19 has the atomic modeset interfaces and helpers, so drivers can now be
45converted over. Modern compositors like Wayland or Surfaceflinger on Android
46really want an atomic modeset interface, so this is all about the bright
47future.
48
49There is a conversion guide for atomic and all you need is a GPU for a
50non-converted driver (again virtual HW drivers for KVM are still all
51suitable).
52
53As part of this drivers also need to convert to universal plane (which means
54exposing primary & cursor as proper plane objects). But that's much easier to
55do by directly using the new atomic helper driver callbacks.
56
57Contact: Daniel Vetter, respective driver maintainers
58
59Level: Advanced
60
61Clean up the clipped coordination confusion around planes
62---------------------------------------------------------
63
64We have a helper to get this right with drm_plane_helper_check_update(), but
65it's not consistently used. This should be fixed, preferrably in the atomic
66helpers (and drivers then moved over to clipped coordinates). Probably the
67helper should also be moved from drm_plane_helper.c to the atomic helpers, to
68avoid confusion - the other helpers in that file are all deprecated legacy
69helpers.
70
71Contact: Ville Syrjälä, Daniel Vetter, driver maintainers
72
73Level: Advanced
74
75Improve plane atomic_check helpers
76----------------------------------
77
78Aside from the clipped coordinates right above there's a few suboptimal things
79with the current helpers:
80
81- drm_plane_helper_funcs->atomic_check gets called for enabled or disabled
82  planes. At best this seems to confuse drivers, worst it means they blow up
83  when the plane is disabled without the CRTC. The only special handling is
84  resetting values in the plane state structures, which instead should be moved
85  into the drm_plane_funcs->atomic_duplicate_state functions.
86
87- Once that's done, helpers could stop calling ->atomic_check for disabled
88  planes.
89
90- Then we could go through all the drivers and remove the more-or-less confused
91  checks for plane_state->fb and plane_state->crtc.
92
93Contact: Daniel Vetter
94
95Level: Advanced
96
97Convert early atomic drivers to async commit helpers
98----------------------------------------------------
99
100For the first year the atomic modeset helpers didn't support asynchronous /
101nonblocking commits, and every driver had to hand-roll them. This is fixed
102now, but there's still a pile of existing drivers that easily could be
103converted over to the new infrastructure.
104
105One issue with the helpers is that they require that drivers handle completion
106events for atomic commits correctly. But fixing these bugs is good anyway.
107
108Somewhat related is the legacy_cursor_update hack, which should be replaced with
109the new atomic_async_check/commit functionality in the helpers in drivers that
110still look at that flag.
111
112Contact: Daniel Vetter, respective driver maintainers
113
114Level: Advanced
115
116Fallout from atomic KMS
117-----------------------
118
119``drm_atomic_helper.c`` provides a batch of functions which implement legacy
120IOCTLs on top of the new atomic driver interface. Which is really nice for
121gradual conversion of drivers, but unfortunately the semantic mismatches are
122a bit too severe. So there's some follow-up work to adjust the function
123interfaces to fix these issues:
124
125* atomic needs the lock acquire context. At the moment that's passed around
126  implicitly with some horrible hacks, and it's also allocate with
127  ``GFP_NOFAIL`` behind the scenes. All legacy paths need to start allocating
128  the acquire context explicitly on stack and then also pass it down into
129  drivers explicitly so that the legacy-on-atomic functions can use them.
130
131  Except for some driver code this is done. This task should be finished by
132  adding WARN_ON(!drm_drv_uses_atomic_modeset) in drm_modeset_lock_all().
133
134* A bunch of the vtable hooks are now in the wrong place: DRM has a split
135  between core vfunc tables (named ``drm_foo_funcs``), which are used to
136  implement the userspace ABI. And then there's the optional hooks for the
137  helper libraries (name ``drm_foo_helper_funcs``), which are purely for
138  internal use. Some of these hooks should be move from ``_funcs`` to
139  ``_helper_funcs`` since they are not part of the core ABI. There's a
140  ``FIXME`` comment in the kerneldoc for each such case in ``drm_crtc.h``.
141
142Contact: Daniel Vetter
143
144Level: Intermediate
145
146Get rid of dev->struct_mutex from GEM drivers
147---------------------------------------------
148
149``dev->struct_mutex`` is the Big DRM Lock from legacy days and infested
150everything. Nowadays in modern drivers the only bit where it's mandatory is
151serializing GEM buffer object destruction. Which unfortunately means drivers
152have to keep track of that lock and either call ``unreference`` or
153``unreference_locked`` depending upon context.
154
155Core GEM doesn't have a need for ``struct_mutex`` any more since kernel 4.8,
156and there's a GEM object ``free`` callback for any drivers which are
157entirely ``struct_mutex`` free.
158
159For drivers that need ``struct_mutex`` it should be replaced with a driver-
160private lock. The tricky part is the BO free functions, since those can't
161reliably take that lock any more. Instead state needs to be protected with
162suitable subordinate locks or some cleanup work pushed to a worker thread. For
163performance-critical drivers it might also be better to go with a more
164fine-grained per-buffer object and per-context lockings scheme. Currently only
165the ``msm`` and `i915` drivers use ``struct_mutex``.
166
167Contact: Daniel Vetter, respective driver maintainers
168
169Level: Advanced
170
171Convert logging to drm_* functions with drm_device paramater
172------------------------------------------------------------
173
174For drivers which could have multiple instances, it is necessary to
175differentiate between which is which in the logs. Since DRM_INFO/WARN/ERROR
176don't do this, drivers used dev_info/warn/err to make this differentiation. We
177now have drm_* variants of the drm print functions, so we can start to convert
178those drivers back to using drm-formatted specific log messages.
179
180Before you start this conversion please contact the relevant maintainers to make
181sure your work will be merged - not everyone agrees that the DRM dmesg macros
182are better.
183
184Contact: Sean Paul, Maintainer of the driver you plan to convert
185
186Level: Starter
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(). Also there's still open-coded version
194of the atomic suspend/resume code in older atomic modeset drivers.
195
196Contact: Maintainer of the driver you plan to convert
197
198Level: Intermediate
199
200Convert drivers to use drm_fbdev_generic_setup()
201------------------------------------------------
202
203Most drivers can use drm_fbdev_generic_setup(). Driver have to implement
204atomic modesetting and GEM vmap support. Current generic fbdev emulation
205expects the framebuffer in system memory (or system-like memory).
206
207Contact: Maintainer of the driver you plan to convert
208
209Level: Intermediate
210
211drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup
212-----------------------------------------------------------------
213
214A lot more drivers could be switched over to the drm_gem_framebuffer helpers.
215Various hold-ups:
216
217- Need to switch over to the generic dirty tracking code using
218  drm_atomic_helper_dirtyfb first (e.g. qxl).
219
220- Need to switch to drm_fbdev_generic_setup(), otherwise a lot of the custom fb
221  setup code can't be deleted.
222
223- Many drivers wrap drm_gem_fb_create() only to check for valid formats. For
224  atomic drivers we could check for valid formats by calling
225  drm_plane_check_pixel_format() against all planes, and pass if any plane
226  supports the format. For non-atomic that's not possible since like the format
227  list for the primary plane is fake and we'd therefor reject valid formats.
228
229- Many drivers subclass drm_framebuffer, we'd need a embedding compatible
230  version of the varios drm_gem_fb_create functions. Maybe called
231  drm_gem_fb_create/_with_dirty/_with_funcs as needed.
232
233Contact: Daniel Vetter
234
235Level: Intermediate
236
237Clean up mmap forwarding
238------------------------
239
240A lot of drivers forward gem mmap calls to dma-buf mmap for imported buffers.
241And also a lot of them forward dma-buf mmap to the gem mmap implementations.
242There's drm_gem_prime_mmap() for this now, but still needs to be rolled out.
243
244Contact: Daniel Vetter
245
246Level: Intermediate
247
248Generic fbdev defio support
249---------------------------
250
251The defio support code in the fbdev core has some very specific requirements,
252which means drivers need to have a special framebuffer for fbdev. The main
253issue is that it uses some fields in struct page itself, which breaks shmem
254gem objects (and other things). To support defio, affected drivers require
255the use of a shadow buffer, which may add CPU and memory overhead.
256
257Possible solution would be to write our own defio mmap code in the drm fbdev
258emulation. It would need to fully wrap the existing mmap ops, forwarding
259everything after it has done the write-protect/mkwrite trickery:
260
261- In the drm_fbdev_fb_mmap helper, if we need defio, change the
262  default page prots to write-protected with something like this::
263
264      vma->vm_page_prot = pgprot_wrprotect(vma->vm_page_prot);
265
266- Set the mkwrite and fsync callbacks with similar implementions to the core
267  fbdev defio stuff. These should all work on plain ptes, they don't actually
268  require a struct page.  uff. These should all work on plain ptes, they don't
269  actually require a struct page.
270
271- Track the dirty pages in a separate structure (bitfield with one bit per page
272  should work) to avoid clobbering struct page.
273
274Might be good to also have some igt testcases for this.
275
276Contact: Daniel Vetter, Noralf Tronnes
277
278Level: Advanced
279
280idr_init_base()
281---------------
282
283DRM core&drivers uses a lot of idr (integer lookup directories) for mapping
284userspace IDs to internal objects, and in most places ID=0 means NULL and hence
285is never used. Switching to idr_init_base() for these would make the idr more
286efficient.
287
288Contact: Daniel Vetter
289
290Level: Starter
291
292struct drm_gem_object_funcs
293---------------------------
294
295GEM objects can now have a function table instead of having the callbacks on the
296DRM driver struct. This is now the preferred way. Callbacks in drivers have been
297converted, except for struct drm_driver.gem_prime_mmap.
298
299Level: Intermediate
300
301Use DRM_MODESET_LOCK_ALL_* helpers instead of boilerplate
302---------------------------------------------------------
303
304For cases where drivers are attempting to grab the modeset locks with a local
305acquire context. Replace the boilerplate code surrounding
306drm_modeset_lock_all_ctx() with DRM_MODESET_LOCK_ALL_BEGIN() and
307DRM_MODESET_LOCK_ALL_END() instead.
308
309This should also be done for all places where drm_modeset_lock_all() is still
310used.
311
312As a reference, take a look at the conversions already completed in drm core.
313
314Contact: Sean Paul, respective driver maintainers
315
316Level: Starter
317
318Rename CMA helpers to DMA helpers
319---------------------------------
320
321CMA (standing for contiguous memory allocator) is really a bit an accident of
322what these were used for first, a much better name would be DMA helpers. In the
323text these should even be called coherent DMA memory helpers (so maybe CDM, but
324no one knows what that means) since underneath they just use dma_alloc_coherent.
325
326Contact: Laurent Pinchart, Daniel Vetter
327
328Level: Intermediate (mostly because it is a huge tasks without good partial
329milestones, not technically itself that challenging)
330
331connector register/unregister fixes
332-----------------------------------
333
334- For most connectors it's a no-op to call drm_connector_register/unregister
335  directly from driver code, drm_dev_register/unregister take care of this
336  already. We can remove all of them.
337
338- For dp drivers it's a bit more a mess, since we need the connector to be
339  registered when calling drm_dp_aux_register. Fix this by instead calling
340  drm_dp_aux_init, and moving the actual registering into a late_register
341  callback as recommended in the kerneldoc.
342
343Level: Intermediate
344
345Remove load/unload callbacks from all non-DRIVER_LEGACY drivers
346---------------------------------------------------------------
347
348The load/unload callbacks in struct &drm_driver are very much midlayers, plus
349for historical reasons they get the ordering wrong (and we can't fix that)
350between setting up the &drm_driver structure and calling drm_dev_register().
351
352- Rework drivers to no longer use the load/unload callbacks, directly coding the
353  load/unload sequence into the driver's probe function.
354
355- Once all non-DRIVER_LEGACY drivers are converted, disallow the load/unload
356  callbacks for all modern drivers.
357
358Contact: Daniel Vetter
359
360Level: Intermediate
361
362Replace drm_detect_hdmi_monitor() with drm_display_info.is_hdmi
363---------------------------------------------------------------
364
365Once EDID is parsed, the monitor HDMI support information is available through
366drm_display_info.is_hdmi. Many drivers still call drm_detect_hdmi_monitor() to
367retrieve the same information, which is less efficient.
368
369Audit each individual driver calling drm_detect_hdmi_monitor() and switch to
370drm_display_info.is_hdmi if applicable.
371
372Contact: Laurent Pinchart, respective driver maintainers
373
374Level: Intermediate
375
376Consolidate custom driver modeset properties
377--------------------------------------------
378
379Before atomic modeset took place, many drivers where creating their own
380properties. Among other things, atomic brought the requirement that custom,
381driver specific properties should not be used.
382
383For this task, we aim to introduce core helpers or reuse the existing ones
384if available:
385
386A quick, unconfirmed, examples list.
387
388Introduce core helpers:
389- audio (amdgpu, intel, gma500, radeon)
390- brightness, contrast, etc (armada, nouveau) - overlay only (?)
391- broadcast rgb (gma500, intel)
392- colorkey (armada, nouveau, rcar) - overlay only (?)
393- dither (amdgpu, nouveau, radeon) - varies across drivers
394- underscan family (amdgpu, radeon, nouveau)
395
396Already in core:
397- colorspace (sti)
398- tv format names, enhancements (gma500, intel)
399- tv overscan, margins, etc. (gma500, intel)
400- zorder (omapdrm) - same as zpos (?)
401
402
403Contact: Emil Velikov, respective driver maintainers
404
405Level: Intermediate
406
407Plumb drm_atomic_state all over
408-------------------------------
409
410Currently various atomic functions take just a single or a handful of
411object states (eg. plane state). While that single object state can
412suffice for some simple cases, we often have to dig out additional
413object states for dealing with various dependencies between the individual
414objects or the hardware they represent. The process of digging out the
415additional states is rather non-intuitive and error prone.
416
417To fix that most functions should rather take the overall
418drm_atomic_state as one of their parameters. The other parameters
419would generally be the object(s) we mainly want to interact with.
420
421For example, instead of
422
423.. code-block:: c
424
425   int (*atomic_check)(struct drm_plane *plane, struct drm_plane_state *state);
426
427we would have something like
428
429.. code-block:: c
430
431   int (*atomic_check)(struct drm_plane *plane, struct drm_atomic_state *state);
432
433The implementation can then trivially gain access to any required object
434state(s) via drm_atomic_get_plane_state(), drm_atomic_get_new_plane_state(),
435drm_atomic_get_old_plane_state(), and their equivalents for
436other object types.
437
438Additionally many drivers currently access the object->state pointer
439directly in their commit functions. That is not going to work if we
440eg. want to allow deeper commit pipelines as those pointers could
441then point to the states corresponding to a future commit instead of
442the current commit we're trying to process. Also non-blocking commits
443execute locklessly so there are serious concerns with dereferencing
444the object->state pointers without holding the locks that protect them.
445Use of drm_atomic_get_new_plane_state(), drm_atomic_get_old_plane_state(),
446etc. avoids these problems as well since they relate to a specific
447commit via the passed in drm_atomic_state.
448
449Contact: Ville Syrjälä, Daniel Vetter
450
451Level: Intermediate
452
453
454Core refactorings
455=================
456
457Make panic handling work
458------------------------
459
460This is a really varied tasks with lots of little bits and pieces:
461
462* The panic path can't be tested currently, leading to constant breaking. The
463  main issue here is that panics can be triggered from hardirq contexts and
464  hence all panic related callback can run in hardirq context. It would be
465  awesome if we could test at least the fbdev helper code and driver code by
466  e.g. trigger calls through drm debugfs files. hardirq context could be
467  achieved by using an IPI to the local processor.
468
469* There's a massive confusion of different panic handlers. DRM fbdev emulation
470  helpers have one, but on top of that the fbcon code itself also has one. We
471  need to make sure that they stop fighting over each another.
472
473* ``drm_can_sleep()`` is a mess. It hides real bugs in normal operations and
474  isn't a full solution for panic paths. We need to make sure that it only
475  returns true if there's a panic going on for real, and fix up all the
476  fallout.
477
478* The panic handler must never sleep, which also means it can't ever
479  ``mutex_lock()``. Also it can't grab any other lock unconditionally, not
480  even spinlocks (because NMI and hardirq can panic too). We need to either
481  make sure to not call such paths, or trylock everything. Really tricky.
482
483* For the above locking troubles reasons it's pretty much impossible to
484  attempt a synchronous modeset from panic handlers. The only thing we could
485  try to achive is an atomic ``set_base`` of the primary plane, and hope that
486  it shows up. Everything else probably needs to be delayed to some worker or
487  something else which happens later on. Otherwise it just kills the box
488  harder, prevent the panic from going out on e.g. netconsole.
489
490* There's also proposal for a simplied DRM console instead of the full-blown
491  fbcon and DRM fbdev emulation. Any kind of panic handling tricks should
492  obviously work for both console, in case we ever get kmslog merged.
493
494Contact: Daniel Vetter
495
496Level: Advanced
497
498Clean up the debugfs support
499----------------------------
500
501There's a bunch of issues with it:
502
503- The drm_info_list ->show() function doesn't even bother to cast to the drm
504  structure for you. This is lazy.
505
506- We probably want to have some support for debugfs files on crtc/connectors and
507  maybe other kms objects directly in core. There's even drm_print support in
508  the funcs for these objects to dump kms state, so it's all there. And then the
509  ->show() functions should obviously give you a pointer to the right object.
510
511- The drm_info_list stuff is centered on drm_minor instead of drm_device. For
512  anything we want to print drm_device (or maybe drm_file) is the right thing.
513
514- The drm_driver->debugfs_init hooks we have is just an artifact of the old
515  midlayered load sequence. DRM debugfs should work more like sysfs, where you
516  can create properties/files for an object anytime you want, and the core
517  takes care of publishing/unpuplishing all the files at register/unregister
518  time. Drivers shouldn't need to worry about these technicalities, and fixing
519  this (together with the drm_minor->drm_device move) would allow us to remove
520  debugfs_init.
521
522Contact: Daniel Vetter
523
524Level: Intermediate
525
526KMS cleanups
527------------
528
529Some of these date from the very introduction of KMS in 2008 ...
530
531- Make ->funcs and ->helper_private vtables optional. There's a bunch of empty
532  function tables in drivers, but before we can remove them we need to make sure
533  that all the users in helpers and drivers do correctly check for a NULL
534  vtable.
535
536- Cleanup up the various ->destroy callbacks. A lot of them just wrapt the
537  drm_*_cleanup implementations and can be removed. Some tack a kfree() at the
538  end, for which we could add drm_*_cleanup_kfree(). And then there's the (for
539  historical reasons) misnamed drm_primary_helper_destroy() function.
540
541Level: Intermediate
542
543Better Testing
544==============
545
546Enable trinity for DRM
547----------------------
548
549And fix up the fallout. Should be really interesting ...
550
551Level: Advanced
552
553Make KMS tests in i-g-t generic
554-------------------------------
555
556The i915 driver team maintains an extensive testsuite for the i915 DRM driver,
557including tons of testcases for corner-cases in the modesetting API. It would
558be awesome if those tests (at least the ones not relying on Intel-specific GEM
559features) could be made to run on any KMS driver.
560
561Basic work to run i-g-t tests on non-i915 is done, what's now missing is mass-
562converting things over. For modeset tests we also first need a bit of
563infrastructure to use dumb buffers for untiled buffers, to be able to run all
564the non-i915 specific modeset tests.
565
566Level: Advanced
567
568Extend virtual test driver (VKMS)
569---------------------------------
570
571See the documentation of :ref:`VKMS <vkms>` for more details. This is an ideal
572internship task, since it only requires a virtual machine and can be sized to
573fit the available time.
574
575Contact: Daniel Vetter
576
577Level: See details
578
579Backlight Refactoring
580---------------------
581
582Backlight drivers have a triple enable/disable state, which is a bit overkill.
583Plan to fix this:
584
5851. Roll out backlight_enable() and backlight_disable() helpers everywhere. This
586   has started already.
5872. In all, only look at one of the three status bits set by the above helpers.
5883. Remove the other two status bits.
589
590Contact: Daniel Vetter
591
592Level: Intermediate
593
594Driver Specific
595===============
596
597AMD DC Display Driver
598---------------------
599
600AMD DC is the display driver for AMD devices starting with Vega. There has been
601a bunch of progress cleaning it up but there's still plenty of work to be done.
602
603See drivers/gpu/drm/amd/display/TODO for tasks.
604
605Contact: Harry Wentland, Alex Deucher
606
607Bootsplash
608==========
609
610There is support in place now for writing internal DRM clients making it
611possible to pick up the bootsplash work that was rejected because it was written
612for fbdev.
613
614- [v6,8/8] drm/client: Hack: Add bootsplash example
615  https://patchwork.freedesktop.org/patch/306579/
616
617- [RFC PATCH v2 00/13] Kernel based bootsplash
618  https://lkml.org/lkml/2017/12/13/764
619
620Contact: Sam Ravnborg
621
622Level: Advanced
623
624Outside DRM
625===========
626
627Convert fbdev drivers to DRM
628----------------------------
629
630There are plenty of fbdev drivers for older hardware. Some hwardware has
631become obsolete, but some still provides good(-enough) framebuffers. The
632drivers that are still useful should be converted to DRM and afterwards
633removed from fbdev.
634
635Very simple fbdev drivers can best be converted by starting with a new
636DRM driver. Simple KMS helpers and SHMEM should be able to handle any
637existing hardware. The new driver's call-back functions are filled from
638existing fbdev code.
639
640More complex fbdev drivers can be refactored step-by-step into a DRM
641driver with the help of the DRM fbconv helpers. [1] These helpers provide
642the transition layer between the DRM core infrastructure and the fbdev
643driver interface. Create a new DRM driver on top of the fbconv helpers,
644copy over the fbdev driver, and hook it up to the DRM code. Examples for
645several fbdev drivers are available at [1] and a tutorial of this process
646available at [2]. The result is a primitive DRM driver that can run X11
647and Weston.
648
649 - [1] https://gitlab.freedesktop.org/tzimmermann/linux/tree/fbconv
650 - [2] https://gitlab.freedesktop.org/tzimmermann/linux/blob/fbconv/drivers/gpu/drm/drm_fbconv_helper.c
651
652Contact: Thomas Zimmermann <tzimmermann@suse.de>
653
654Level: Advanced
655