xref: /openbmc/linux/Documentation/gpu/todo.rst (revision 3cea11cd)
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34implementation (there's lots of outdated locking leftovers in various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
108Contact: Daniel Vetter, respective driver maintainers
109
110Level: Advanced
111
112Fallout from atomic KMS
113-----------------------
114
115``drm_atomic_helper.c`` provides a batch of functions which implement legacy
116IOCTLs on top of the new atomic driver interface. Which is really nice for
117gradual conversion of drivers, but unfortunately the semantic mismatches are
118a bit too severe. So there's some follow-up work to adjust the function
119interfaces to fix these issues:
120
121* atomic needs the lock acquire context. At the moment that's passed around
122  implicitly with some horrible hacks, and it's also allocate with
123  ``GFP_NOFAIL`` behind the scenes. All legacy paths need to start allocating
124  the acquire context explicitly on stack and then also pass it down into
125  drivers explicitly so that the legacy-on-atomic functions can use them.
126
127  Except for some driver code this is done. This task should be finished by
128  adding WARN_ON(!drm_drv_uses_atomic_modeset) in drm_modeset_lock_all().
129
130* A bunch of the vtable hooks are now in the wrong place: DRM has a split
131  between core vfunc tables (named ``drm_foo_funcs``), which are used to
132  implement the userspace ABI. And then there's the optional hooks for the
133  helper libraries (name ``drm_foo_helper_funcs``), which are purely for
134  internal use. Some of these hooks should be move from ``_funcs`` to
135  ``_helper_funcs`` since they are not part of the core ABI. There's a
136  ``FIXME`` comment in the kerneldoc for each such case in ``drm_crtc.h``.
137
138Contact: Daniel Vetter
139
140Level: Intermediate
141
142Get rid of dev->struct_mutex from GEM drivers
143---------------------------------------------
144
145``dev->struct_mutex`` is the Big DRM Lock from legacy days and infested
146everything. Nowadays in modern drivers the only bit where it's mandatory is
147serializing GEM buffer object destruction. Which unfortunately means drivers
148have to keep track of that lock and either call ``unreference`` or
149``unreference_locked`` depending upon context.
150
151Core GEM doesn't have a need for ``struct_mutex`` any more since kernel 4.8,
152and there's a ``gem_free_object_unlocked`` callback for any drivers which are
153entirely ``struct_mutex`` free.
154
155For drivers that need ``struct_mutex`` it should be replaced with a driver-
156private lock. The tricky part is the BO free functions, since those can't
157reliably take that lock any more. Instead state needs to be protected with
158suitable subordinate locks or some cleanup work pushed to a worker thread. For
159performance-critical drivers it might also be better to go with a more
160fine-grained per-buffer object and per-context lockings scheme. Currently only
161the ``msm`` and `i915` drivers use ``struct_mutex``.
162
163Contact: Daniel Vetter, respective driver maintainers
164
165Level: Advanced
166
167Convert logging to drm_* functions with drm_device paramater
168------------------------------------------------------------
169
170For drivers which could have multiple instances, it is necessary to
171differentiate between which is which in the logs. Since DRM_INFO/WARN/ERROR
172don't do this, drivers used dev_info/warn/err to make this differentiation. We
173now have drm_* variants of the drm print functions, so we can start to convert
174those drivers back to using drm-formatted specific log messages.
175
176Before you start this conversion please contact the relevant maintainers to make
177sure your work will be merged - not everyone agrees that the DRM dmesg macros
178are better.
179
180Contact: Sean Paul, Maintainer of the driver you plan to convert
181
182Level: Starter
183
184Convert drivers to use simple modeset suspend/resume
185----------------------------------------------------
186
187Most drivers (except i915 and nouveau) that use
188drm_atomic_helper_suspend/resume() can probably be converted to use
189drm_mode_config_helper_suspend/resume(). Also there's still open-coded version
190of the atomic suspend/resume code in older atomic modeset drivers.
191
192Contact: Maintainer of the driver you plan to convert
193
194Level: Intermediate
195
196Convert drivers to use drm_fbdev_generic_setup()
197------------------------------------------------
198
199Most drivers can use drm_fbdev_generic_setup(). Driver have to implement
200atomic modesetting and GEM vmap support. Current generic fbdev emulation
201expects the framebuffer in system memory (or system-like memory).
202
203Contact: Maintainer of the driver you plan to convert
204
205Level: Intermediate
206
207drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup
208-----------------------------------------------------------------
209
210A lot more drivers could be switched over to the drm_gem_framebuffer helpers.
211Various hold-ups:
212
213- Need to switch over to the generic dirty tracking code using
214  drm_atomic_helper_dirtyfb first (e.g. qxl).
215
216- Need to switch to drm_fbdev_generic_setup(), otherwise a lot of the custom fb
217  setup code can't be deleted.
218
219- Many drivers wrap drm_gem_fb_create() only to check for valid formats. For
220  atomic drivers we could check for valid formats by calling
221  drm_plane_check_pixel_format() against all planes, and pass if any plane
222  supports the format. For non-atomic that's not possible since like the format
223  list for the primary plane is fake and we'd therefor reject valid formats.
224
225- Many drivers subclass drm_framebuffer, we'd need a embedding compatible
226  version of the varios drm_gem_fb_create functions. Maybe called
227  drm_gem_fb_create/_with_dirty/_with_funcs as needed.
228
229Contact: Daniel Vetter
230
231Level: Intermediate
232
233Clean up mmap forwarding
234------------------------
235
236A lot of drivers forward gem mmap calls to dma-buf mmap for imported buffers.
237And also a lot of them forward dma-buf mmap to the gem mmap implementations.
238There's drm_gem_prime_mmap() for this now, but still needs to be rolled out.
239
240Contact: Daniel Vetter
241
242Level: Intermediate
243
244Generic fbdev defio support
245---------------------------
246
247The defio support code in the fbdev core has some very specific requirements,
248which means drivers need to have a special framebuffer for fbdev. The main
249issue is that it uses some fields in struct page itself, which breaks shmem
250gem objects (and other things). To support defio, affected drivers require
251the use of a shadow buffer, which may add CPU and memory overhead.
252
253Possible solution would be to write our own defio mmap code in the drm fbdev
254emulation. It would need to fully wrap the existing mmap ops, forwarding
255everything after it has done the write-protect/mkwrite trickery:
256
257- In the drm_fbdev_fb_mmap helper, if we need defio, change the
258  default page prots to write-protected with something like this::
259
260      vma->vm_page_prot = pgprot_wrprotect(vma->vm_page_prot);
261
262- Set the mkwrite and fsync callbacks with similar implementions to the core
263  fbdev defio stuff. These should all work on plain ptes, they don't actually
264  require a struct page.  uff. These should all work on plain ptes, they don't
265  actually require a struct page.
266
267- Track the dirty pages in a separate structure (bitfield with one bit per page
268  should work) to avoid clobbering struct page.
269
270Might be good to also have some igt testcases for this.
271
272Contact: Daniel Vetter, Noralf Tronnes
273
274Level: Advanced
275
276idr_init_base()
277---------------
278
279DRM core&drivers uses a lot of idr (integer lookup directories) for mapping
280userspace IDs to internal objects, and in most places ID=0 means NULL and hence
281is never used. Switching to idr_init_base() for these would make the idr more
282efficient.
283
284Contact: Daniel Vetter
285
286Level: Starter
287
288struct drm_gem_object_funcs
289---------------------------
290
291GEM objects can now have a function table instead of having the callbacks on the
292DRM driver struct. This is now the preferred way and drivers can be moved over.
293
294We also need a 2nd version of the CMA define that doesn't require the
295vmapping to be present (different hook for prime importing). Plus this needs to
296be rolled out to all drivers using their own implementations, too.
297
298Level: Intermediate
299
300Use DRM_MODESET_LOCK_ALL_* helpers instead of boilerplate
301---------------------------------------------------------
302
303For cases where drivers are attempting to grab the modeset locks with a local
304acquire context. Replace the boilerplate code surrounding
305drm_modeset_lock_all_ctx() with DRM_MODESET_LOCK_ALL_BEGIN() and
306DRM_MODESET_LOCK_ALL_END() instead.
307
308This should also be done for all places where drm_modeset_lock_all() is still
309used.
310
311As a reference, take a look at the conversions already completed in drm core.
312
313Contact: Sean Paul, respective driver maintainers
314
315Level: Starter
316
317Rename CMA helpers to DMA helpers
318---------------------------------
319
320CMA (standing for contiguous memory allocator) is really a bit an accident of
321what these were used for first, a much better name would be DMA helpers. In the
322text these should even be called coherent DMA memory helpers (so maybe CDM, but
323no one knows what that means) since underneath they just use dma_alloc_coherent.
324
325Contact: Laurent Pinchart, Daniel Vetter
326
327Level: Intermediate (mostly because it is a huge tasks without good partial
328milestones, not technically itself that challenging)
329
330connector register/unregister fixes
331-----------------------------------
332
333- For most connectors it's a no-op to call drm_connector_register/unregister
334  directly from driver code, drm_dev_register/unregister take care of this
335  already. We can remove all of them.
336
337- For dp drivers it's a bit more a mess, since we need the connector to be
338  registered when calling drm_dp_aux_register. Fix this by instead calling
339  drm_dp_aux_init, and moving the actual registering into a late_register
340  callback as recommended in the kerneldoc.
341
342Level: Intermediate
343
344Remove load/unload callbacks from all non-DRIVER_LEGACY drivers
345---------------------------------------------------------------
346
347The load/unload callbacks in struct &drm_driver are very much midlayers, plus
348for historical reasons they get the ordering wrong (and we can't fix that)
349between setting up the &drm_driver structure and calling drm_dev_register().
350
351- Rework drivers to no longer use the load/unload callbacks, directly coding the
352  load/unload sequence into the driver's probe function.
353
354- Once all non-DRIVER_LEGACY drivers are converted, disallow the load/unload
355  callbacks for all modern drivers.
356
357Contact: Daniel Vetter
358
359Level: Intermediate
360
361Replace drm_detect_hdmi_monitor() with drm_display_info.is_hdmi
362---------------------------------------------------------------
363
364Once EDID is parsed, the monitor HDMI support information is available through
365drm_display_info.is_hdmi. Many drivers still call drm_detect_hdmi_monitor() to
366retrieve the same information, which is less efficient.
367
368Audit each individual driver calling drm_detect_hdmi_monitor() and switch to
369drm_display_info.is_hdmi if applicable.
370
371Contact: Laurent Pinchart, respective driver maintainers
372
373Level: Intermediate
374
375Consolidate custom driver modeset properties
376--------------------------------------------
377
378Before atomic modeset took place, many drivers where creating their own
379properties. Among other things, atomic brought the requirement that custom,
380driver specific properties should not be used.
381
382For this task, we aim to introduce core helpers or reuse the existing ones
383if available:
384
385A quick, unconfirmed, examples list.
386
387Introduce core helpers:
388- audio (amdgpu, intel, gma500, radeon)
389- brightness, contrast, etc (armada, nouveau) - overlay only (?)
390- broadcast rgb (gma500, intel)
391- colorkey (armada, nouveau, rcar) - overlay only (?)
392- dither (amdgpu, nouveau, radeon) - varies across drivers
393- underscan family (amdgpu, radeon, nouveau)
394
395Already in core:
396- colorspace (sti)
397- tv format names, enhancements (gma500, intel)
398- tv overscan, margins, etc. (gma500, intel)
399- zorder (omapdrm) - same as zpos (?)
400
401
402Contact: Emil Velikov, respective driver maintainers
403
404Level: Intermediate
405
406Plumb drm_atomic_state all over
407-------------------------------
408
409Currently various atomic functions take just a single or a handful of
410object states (eg. plane state). While that single object state can
411suffice for some simple cases, we often have to dig out additional
412object states for dealing with various dependencies between the individual
413objects or the hardware they represent. The process of digging out the
414additional states is rather non-intuitive and error prone.
415
416To fix that most functions should rather take the overall
417drm_atomic_state as one of their parameters. The other parameters
418would generally be the object(s) we mainly want to interact with.
419
420For example, instead of
421
422.. code-block:: c
423
424   int (*atomic_check)(struct drm_plane *plane, struct drm_plane_state *state);
425
426we would have something like
427
428.. code-block:: c
429
430   int (*atomic_check)(struct drm_plane *plane, struct drm_atomic_state *state);
431
432The implementation can then trivially gain access to any required object
433state(s) via drm_atomic_get_plane_state(), drm_atomic_get_new_plane_state(),
434drm_atomic_get_old_plane_state(), and their equivalents for
435other object types.
436
437Additionally many drivers currently access the object->state pointer
438directly in their commit functions. That is not going to work if we
439eg. want to allow deeper commit pipelines as those pointers could
440then point to the states corresponding to a future commit instead of
441the current commit we're trying to process. Also non-blocking commits
442execute locklessly so there are serious concerns with dereferencing
443the object->state pointers without holding the locks that protect them.
444Use of drm_atomic_get_new_plane_state(), drm_atomic_get_old_plane_state(),
445etc. avoids these problems as well since they relate to a specific
446commit via the passed in drm_atomic_state.
447
448Contact: Ville Syrjälä, Daniel Vetter
449
450Level: Intermediate
451
452
453Core refactorings
454=================
455
456Make panic handling work
457------------------------
458
459This is a really varied tasks with lots of little bits and pieces:
460
461* The panic path can't be tested currently, leading to constant breaking. The
462  main issue here is that panics can be triggered from hardirq contexts and
463  hence all panic related callback can run in hardirq context. It would be
464  awesome if we could test at least the fbdev helper code and driver code by
465  e.g. trigger calls through drm debugfs files. hardirq context could be
466  achieved by using an IPI to the local processor.
467
468* There's a massive confusion of different panic handlers. DRM fbdev emulation
469  helpers have one, but on top of that the fbcon code itself also has one. We
470  need to make sure that they stop fighting over each another.
471
472* ``drm_can_sleep()`` is a mess. It hides real bugs in normal operations and
473  isn't a full solution for panic paths. We need to make sure that it only
474  returns true if there's a panic going on for real, and fix up all the
475  fallout.
476
477* The panic handler must never sleep, which also means it can't ever
478  ``mutex_lock()``. Also it can't grab any other lock unconditionally, not
479  even spinlocks (because NMI and hardirq can panic too). We need to either
480  make sure to not call such paths, or trylock everything. Really tricky.
481
482* For the above locking troubles reasons it's pretty much impossible to
483  attempt a synchronous modeset from panic handlers. The only thing we could
484  try to achive is an atomic ``set_base`` of the primary plane, and hope that
485  it shows up. Everything else probably needs to be delayed to some worker or
486  something else which happens later on. Otherwise it just kills the box
487  harder, prevent the panic from going out on e.g. netconsole.
488
489* There's also proposal for a simplied DRM console instead of the full-blown
490  fbcon and DRM fbdev emulation. Any kind of panic handling tricks should
491  obviously work for both console, in case we ever get kmslog merged.
492
493Contact: Daniel Vetter
494
495Level: Advanced
496
497Clean up the debugfs support
498----------------------------
499
500There's a bunch of issues with it:
501
502- The drm_info_list ->show() function doesn't even bother to cast to the drm
503  structure for you. This is lazy.
504
505- We probably want to have some support for debugfs files on crtc/connectors and
506  maybe other kms objects directly in core. There's even drm_print support in
507  the funcs for these objects to dump kms state, so it's all there. And then the
508  ->show() functions should obviously give you a pointer to the right object.
509
510- The drm_info_list stuff is centered on drm_minor instead of drm_device. For
511  anything we want to print drm_device (or maybe drm_file) is the right thing.
512
513- The drm_driver->debugfs_init hooks we have is just an artifact of the old
514  midlayered load sequence. DRM debugfs should work more like sysfs, where you
515  can create properties/files for an object anytime you want, and the core
516  takes care of publishing/unpuplishing all the files at register/unregister
517  time. Drivers shouldn't need to worry about these technicalities, and fixing
518  this (together with the drm_minor->drm_device move) would allow us to remove
519  debugfs_init.
520
521- Drop the return code and error checking from all debugfs functions. Greg KH is
522  working on this already.
523
524Contact: Daniel Vetter
525
526Level: Intermediate
527
528KMS cleanups
529------------
530
531Some of these date from the very introduction of KMS in 2008 ...
532
533- Make ->funcs and ->helper_private vtables optional. There's a bunch of empty
534  function tables in drivers, but before we can remove them we need to make sure
535  that all the users in helpers and drivers do correctly check for a NULL
536  vtable.
537
538- Cleanup up the various ->destroy callbacks. A lot of them just wrapt the
539  drm_*_cleanup implementations and can be removed. Some tack a kfree() at the
540  end, for which we could add drm_*_cleanup_kfree(). And then there's the (for
541  historical reasons) misnamed drm_primary_helper_destroy() function.
542
543Level: Intermediate
544
545Better Testing
546==============
547
548Enable trinity for DRM
549----------------------
550
551And fix up the fallout. Should be really interesting ...
552
553Level: Advanced
554
555Make KMS tests in i-g-t generic
556-------------------------------
557
558The i915 driver team maintains an extensive testsuite for the i915 DRM driver,
559including tons of testcases for corner-cases in the modesetting API. It would
560be awesome if those tests (at least the ones not relying on Intel-specific GEM
561features) could be made to run on any KMS driver.
562
563Basic work to run i-g-t tests on non-i915 is done, what's now missing is mass-
564converting things over. For modeset tests we also first need a bit of
565infrastructure to use dumb buffers for untiled buffers, to be able to run all
566the non-i915 specific modeset tests.
567
568Level: Advanced
569
570Extend virtual test driver (VKMS)
571---------------------------------
572
573See the documentation of :ref:`VKMS <vkms>` for more details. This is an ideal
574internship task, since it only requires a virtual machine and can be sized to
575fit the available time.
576
577Contact: Daniel Vetter
578
579Level: See details
580
581Backlight Refactoring
582---------------------
583
584Backlight drivers have a triple enable/disable state, which is a bit overkill.
585Plan to fix this:
586
5871. Roll out backlight_enable() and backlight_disable() helpers everywhere. This
588   has started already.
5892. In all, only look at one of the three status bits set by the above helpers.
5903. Remove the other two status bits.
591
592Contact: Daniel Vetter
593
594Level: Intermediate
595
596Driver Specific
597===============
598
599AMD DC Display Driver
600---------------------
601
602AMD DC is the display driver for AMD devices starting with Vega. There has been
603a bunch of progress cleaning it up but there's still plenty of work to be done.
604
605See drivers/gpu/drm/amd/display/TODO for tasks.
606
607Contact: Harry Wentland, Alex Deucher
608
609Bootsplash
610==========
611
612There is support in place now for writing internal DRM clients making it
613possible to pick up the bootsplash work that was rejected because it was written
614for fbdev.
615
616- [v6,8/8] drm/client: Hack: Add bootsplash example
617  https://patchwork.freedesktop.org/patch/306579/
618
619- [RFC PATCH v2 00/13] Kernel based bootsplash
620  https://lkml.org/lkml/2017/12/13/764
621
622Contact: Sam Ravnborg
623
624Level: Advanced
625
626Outside DRM
627===========
628
629Convert fbdev drivers to DRM
630----------------------------
631
632There are plenty of fbdev drivers for older hardware. Some hwardware has
633become obsolete, but some still provides good(-enough) framebuffers. The
634drivers that are still useful should be converted to DRM and afterwards
635removed from fbdev.
636
637Very simple fbdev drivers can best be converted by starting with a new
638DRM driver. Simple KMS helpers and SHMEM should be able to handle any
639existing hardware. The new driver's call-back functions are filled from
640existing fbdev code.
641
642More complex fbdev drivers can be refactored step-by-step into a DRM
643driver with the help of the DRM fbconv helpers. [1] These helpers provide
644the transition layer between the DRM core infrastructure and the fbdev
645driver interface. Create a new DRM driver on top of the fbconv helpers,
646copy over the fbdev driver, and hook it up to the DRM code. Examples for
647several fbdev drivers are available at [1] and a tutorial of this process
648available at [2]. The result is a primitive DRM driver that can run X11
649and Weston.
650
651 - [1] https://gitlab.freedesktop.org/tzimmermann/linux/tree/fbconv
652 - [2] https://gitlab.freedesktop.org/tzimmermann/linux/blob/fbconv/drivers/gpu/drm/drm_fbconv_helper.c
653
654Contact: Thomas Zimmermann <tzimmermann@suse.de>
655
656Level: Advanced
657