7 This section contains a list of smaller janitorial tasks in the kernel DRM
8 graphics subsystem useful as newbie projects. Or for slow rainy days.
13 To make it easier task are categorized into different levels:
15 Starter: Good tasks to get started with the DRM subsystem.
17 Intermediate: Tasks which need some experience with working in the DRM
18 subsystem, or some specific GPU/display graphics knowledge. For debugging issue
19 it's good to have the relevant hardware (or a virtual driver set up) available
22 Advanced: Tricky tasks that need fairly good understanding of the DRM subsystem
23 and graphics topics. Generally need the relevant hardware for development and
26 Expert: Only attempt these if you've successfully completed some tricky
27 refactorings already and are an expert in the specific area
29 Subsystem-wide refactorings
30 ===========================
32 Remove custom dumb_map_offset implementations
33 ---------------------------------------------
35 All GEM based drivers should be using drm_gem_create_mmap_offset() instead.
36 Audit each individual driver, make sure it'll work with the generic
37 implementation (there's lots of outdated locking leftovers in various
38 implementations), and then remove it.
40 Contact: Daniel Vetter, respective driver maintainers
44 Convert existing KMS drivers to atomic modesetting
45 --------------------------------------------------
47 3.19 has the atomic modeset interfaces and helpers, so drivers can now be
48 converted over. Modern compositors like Wayland or Surfaceflinger on Android
49 really want an atomic modeset interface, so this is all about the bright
52 There is a conversion guide for atomic and all you need is a GPU for a
53 non-converted driver (again virtual HW drivers for KVM are still all
56 As part of this drivers also need to convert to universal plane (which means
57 exposing primary & cursor as proper plane objects). But that's much easier to
58 do by directly using the new atomic helper driver callbacks.
60 Contact: Daniel Vetter, respective driver maintainers
64 Clean up the clipped coordination confusion around planes
65 ---------------------------------------------------------
67 We have a helper to get this right with drm_plane_helper_check_update(), but
68 it's not consistently used. This should be fixed, preferrably in the atomic
69 helpers (and drivers then moved over to clipped coordinates). Probably the
70 helper should also be moved from drm_plane_helper.c to the atomic helpers, to
71 avoid confusion - the other helpers in that file are all deprecated legacy
74 Contact: Ville Syrjälä, Daniel Vetter, driver maintainers
78 Improve plane atomic_check helpers
79 ----------------------------------
81 Aside from the clipped coordinates right above there's a few suboptimal things
82 with the current helpers:
84 - drm_plane_helper_funcs->atomic_check gets called for enabled or disabled
85 planes. At best this seems to confuse drivers, worst it means they blow up
86 when the plane is disabled without the CRTC. The only special handling is
87 resetting values in the plane state structures, which instead should be moved
88 into the drm_plane_funcs->atomic_duplicate_state functions.
90 - Once that's done, helpers could stop calling ->atomic_check for disabled
93 - Then we could go through all the drivers and remove the more-or-less confused
94 checks for plane_state->fb and plane_state->crtc.
96 Contact: Daniel Vetter
100 Convert early atomic drivers to async commit helpers
101 ----------------------------------------------------
103 For the first year the atomic modeset helpers didn't support asynchronous /
104 nonblocking commits, and every driver had to hand-roll them. This is fixed
105 now, but there's still a pile of existing drivers that easily could be
106 converted over to the new infrastructure.
108 One issue with the helpers is that they require that drivers handle completion
109 events for atomic commits correctly. But fixing these bugs is good anyway.
111 Somewhat related is the legacy_cursor_update hack, which should be replaced with
112 the new atomic_async_check/commit functionality in the helpers in drivers that
113 still look at that flag.
115 Contact: Daniel Vetter, respective driver maintainers
119 Fallout from atomic KMS
120 -----------------------
122 ``drm_atomic_helper.c`` provides a batch of functions which implement legacy
123 IOCTLs on top of the new atomic driver interface. Which is really nice for
124 gradual conversion of drivers, but unfortunately the semantic mismatches are
125 a bit too severe. So there's some follow-up work to adjust the function
126 interfaces to fix these issues:
128 * atomic needs the lock acquire context. At the moment that's passed around
129 implicitly with some horrible hacks, and it's also allocate with
130 ``GFP_NOFAIL`` behind the scenes. All legacy paths need to start allocating
131 the acquire context explicitly on stack and then also pass it down into
132 drivers explicitly so that the legacy-on-atomic functions can use them.
134 Except for some driver code this is done. This task should be finished by
135 adding WARN_ON(!drm_drv_uses_atomic_modeset) in drm_modeset_lock_all().
137 * A bunch of the vtable hooks are now in the wrong place: DRM has a split
138 between core vfunc tables (named ``drm_foo_funcs``), which are used to
139 implement the userspace ABI. And then there's the optional hooks for the
140 helper libraries (name ``drm_foo_helper_funcs``), which are purely for
141 internal use. Some of these hooks should be move from ``_funcs`` to
142 ``_helper_funcs`` since they are not part of the core ABI. There's a
143 ``FIXME`` comment in the kerneldoc for each such case in ``drm_crtc.h``.
145 Contact: Daniel Vetter
149 Get rid of dev->struct_mutex from GEM drivers
150 ---------------------------------------------
152 ``dev->struct_mutex`` is the Big DRM Lock from legacy days and infested
153 everything. Nowadays in modern drivers the only bit where it's mandatory is
154 serializing GEM buffer object destruction. Which unfortunately means drivers
155 have to keep track of that lock and either call ``unreference`` or
156 ``unreference_locked`` depending upon context.
158 Core GEM doesn't have a need for ``struct_mutex`` any more since kernel 4.8,
159 and there's a GEM object ``free`` callback for any drivers which are
160 entirely ``struct_mutex`` free.
162 For drivers that need ``struct_mutex`` it should be replaced with a driver-
163 private lock. The tricky part is the BO free functions, since those can't
164 reliably take that lock any more. Instead state needs to be protected with
165 suitable subordinate locks or some cleanup work pushed to a worker thread. For
166 performance-critical drivers it might also be better to go with a more
167 fine-grained per-buffer object and per-context lockings scheme. Currently only
168 the ``msm`` and `i915` drivers use ``struct_mutex``.
170 Contact: Daniel Vetter, respective driver maintainers
174 Move Buffer Object Locking to dma_resv_lock()
175 ---------------------------------------------
177 Many drivers have their own per-object locking scheme, usually using
178 mutex_lock(). This causes all kinds of trouble for buffer sharing, since
179 depending which driver is the exporter and importer, the locking hierarchy is
182 To solve this we need one standard per-object locking mechanism, which is
183 dma_resv_lock(). This lock needs to be called as the outermost lock, with all
184 other driver specific per-object locks removed. The problem is tha rolling out
185 the actual change to the locking contract is a flag day, due to struct dma_buf
190 Convert logging to drm_* functions with drm_device paramater
191 ------------------------------------------------------------
193 For drivers which could have multiple instances, it is necessary to
194 differentiate between which is which in the logs. Since DRM_INFO/WARN/ERROR
195 don't do this, drivers used dev_info/warn/err to make this differentiation. We
196 now have drm_* variants of the drm print functions, so we can start to convert
197 those drivers back to using drm-formatted specific log messages.
199 Before you start this conversion please contact the relevant maintainers to make
200 sure your work will be merged - not everyone agrees that the DRM dmesg macros
203 Contact: Sean Paul, Maintainer of the driver you plan to convert
207 Convert drivers to use simple modeset suspend/resume
208 ----------------------------------------------------
210 Most drivers (except i915 and nouveau) that use
211 drm_atomic_helper_suspend/resume() can probably be converted to use
212 drm_mode_config_helper_suspend/resume(). Also there's still open-coded version
213 of the atomic suspend/resume code in older atomic modeset drivers.
215 Contact: Maintainer of the driver you plan to convert
219 Convert drivers to use drm_fbdev_generic_setup()
220 ------------------------------------------------
222 Most drivers can use drm_fbdev_generic_setup(). Driver have to implement
223 atomic modesetting and GEM vmap support. Historically, generic fbdev emulation
224 expected the framebuffer in system memory or system-like memory. By employing
225 struct dma_buf_map, drivers with frambuffers in I/O memory can be supported
228 Contact: Maintainer of the driver you plan to convert
232 Reimplement functions in drm_fbdev_fb_ops without fbdev
233 -------------------------------------------------------
235 A number of callback functions in drm_fbdev_fb_ops could benefit from
236 being rewritten without dependencies on the fbdev module. Some of the
237 helpers could further benefit from using struct dma_buf_map instead of
245 drm_framebuffer_funcs and drm_mode_config_funcs.fb_create cleanup
246 -----------------------------------------------------------------
248 A lot more drivers could be switched over to the drm_gem_framebuffer helpers.
251 - Need to switch over to the generic dirty tracking code using
252 drm_atomic_helper_dirtyfb first (e.g. qxl).
254 - Need to switch to drm_fbdev_generic_setup(), otherwise a lot of the custom fb
255 setup code can't be deleted.
257 - Many drivers wrap drm_gem_fb_create() only to check for valid formats. For
258 atomic drivers we could check for valid formats by calling
259 drm_plane_check_pixel_format() against all planes, and pass if any plane
260 supports the format. For non-atomic that's not possible since like the format
261 list for the primary plane is fake and we'd therefor reject valid formats.
263 - Many drivers subclass drm_framebuffer, we'd need a embedding compatible
264 version of the varios drm_gem_fb_create functions. Maybe called
265 drm_gem_fb_create/_with_dirty/_with_funcs as needed.
267 Contact: Daniel Vetter
271 Clean up mmap forwarding
272 ------------------------
274 A lot of drivers forward gem mmap calls to dma-buf mmap for imported buffers.
275 And also a lot of them forward dma-buf mmap to the gem mmap implementations.
276 There's drm_gem_prime_mmap() for this now, but still needs to be rolled out.
278 Contact: Daniel Vetter
282 Generic fbdev defio support
283 ---------------------------
285 The defio support code in the fbdev core has some very specific requirements,
286 which means drivers need to have a special framebuffer for fbdev. The main
287 issue is that it uses some fields in struct page itself, which breaks shmem
288 gem objects (and other things). To support defio, affected drivers require
289 the use of a shadow buffer, which may add CPU and memory overhead.
291 Possible solution would be to write our own defio mmap code in the drm fbdev
292 emulation. It would need to fully wrap the existing mmap ops, forwarding
293 everything after it has done the write-protect/mkwrite trickery:
295 - In the drm_fbdev_fb_mmap helper, if we need defio, change the
296 default page prots to write-protected with something like this::
298 vma->vm_page_prot = pgprot_wrprotect(vma->vm_page_prot);
300 - Set the mkwrite and fsync callbacks with similar implementions to the core
301 fbdev defio stuff. These should all work on plain ptes, they don't actually
302 require a struct page. uff. These should all work on plain ptes, they don't
303 actually require a struct page.
305 - Track the dirty pages in a separate structure (bitfield with one bit per page
306 should work) to avoid clobbering struct page.
308 Might be good to also have some igt testcases for this.
310 Contact: Daniel Vetter, Noralf Tronnes
314 Garbage collect fbdev scrolling acceleration
315 --------------------------------------------
317 Scroll acceleration is disabled in fbcon by hard-wiring p->scrollmode =
318 SCROLL_REDRAW. There's a ton of code this will allow us to remove:
320 - lots of code in fbcon.c
322 - a bunch of the hooks in fbcon_ops, maybe the remaining hooks could be called
323 directly instead of the function table (with a switch on p->rotate)
325 - fb_copyarea is unused after this, and can be deleted from all drivers
327 Note that not all acceleration code can be deleted, since clearing and cursor
328 support is still accelerated, which might be good candidates for further
331 Contact: Daniel Vetter
338 DRM core&drivers uses a lot of idr (integer lookup directories) for mapping
339 userspace IDs to internal objects, and in most places ID=0 means NULL and hence
340 is never used. Switching to idr_init_base() for these would make the idr more
343 Contact: Daniel Vetter
347 struct drm_gem_object_funcs
348 ---------------------------
350 GEM objects can now have a function table instead of having the callbacks on the
351 DRM driver struct. This is now the preferred way. Callbacks in drivers have been
352 converted, except for struct drm_driver.gem_prime_mmap.
356 Use DRM_MODESET_LOCK_ALL_* helpers instead of boilerplate
357 ---------------------------------------------------------
359 For cases where drivers are attempting to grab the modeset locks with a local
360 acquire context. Replace the boilerplate code surrounding
361 drm_modeset_lock_all_ctx() with DRM_MODESET_LOCK_ALL_BEGIN() and
362 DRM_MODESET_LOCK_ALL_END() instead.
364 This should also be done for all places where drm_modeset_lock_all() is still
367 As a reference, take a look at the conversions already completed in drm core.
369 Contact: Sean Paul, respective driver maintainers
373 Rename CMA helpers to DMA helpers
374 ---------------------------------
376 CMA (standing for contiguous memory allocator) is really a bit an accident of
377 what these were used for first, a much better name would be DMA helpers. In the
378 text these should even be called coherent DMA memory helpers (so maybe CDM, but
379 no one knows what that means) since underneath they just use dma_alloc_coherent.
381 Contact: Laurent Pinchart, Daniel Vetter
383 Level: Intermediate (mostly because it is a huge tasks without good partial
384 milestones, not technically itself that challenging)
386 connector register/unregister fixes
387 -----------------------------------
389 - For most connectors it's a no-op to call drm_connector_register/unregister
390 directly from driver code, drm_dev_register/unregister take care of this
391 already. We can remove all of them.
393 - For dp drivers it's a bit more a mess, since we need the connector to be
394 registered when calling drm_dp_aux_register. Fix this by instead calling
395 drm_dp_aux_init, and moving the actual registering into a late_register
396 callback as recommended in the kerneldoc.
400 Remove load/unload callbacks from all non-DRIVER_LEGACY drivers
401 ---------------------------------------------------------------
403 The load/unload callbacks in struct &drm_driver are very much midlayers, plus
404 for historical reasons they get the ordering wrong (and we can't fix that)
405 between setting up the &drm_driver structure and calling drm_dev_register().
407 - Rework drivers to no longer use the load/unload callbacks, directly coding the
408 load/unload sequence into the driver's probe function.
410 - Once all non-DRIVER_LEGACY drivers are converted, disallow the load/unload
411 callbacks for all modern drivers.
413 Contact: Daniel Vetter
417 Replace drm_detect_hdmi_monitor() with drm_display_info.is_hdmi
418 ---------------------------------------------------------------
420 Once EDID is parsed, the monitor HDMI support information is available through
421 drm_display_info.is_hdmi. Many drivers still call drm_detect_hdmi_monitor() to
422 retrieve the same information, which is less efficient.
424 Audit each individual driver calling drm_detect_hdmi_monitor() and switch to
425 drm_display_info.is_hdmi if applicable.
427 Contact: Laurent Pinchart, respective driver maintainers
431 Consolidate custom driver modeset properties
432 --------------------------------------------
434 Before atomic modeset took place, many drivers where creating their own
435 properties. Among other things, atomic brought the requirement that custom,
436 driver specific properties should not be used.
438 For this task, we aim to introduce core helpers or reuse the existing ones
441 A quick, unconfirmed, examples list.
443 Introduce core helpers:
444 - audio (amdgpu, intel, gma500, radeon)
445 - brightness, contrast, etc (armada, nouveau) - overlay only (?)
446 - broadcast rgb (gma500, intel)
447 - colorkey (armada, nouveau, rcar) - overlay only (?)
448 - dither (amdgpu, nouveau, radeon) - varies across drivers
449 - underscan family (amdgpu, radeon, nouveau)
453 - tv format names, enhancements (gma500, intel)
454 - tv overscan, margins, etc. (gma500, intel)
455 - zorder (omapdrm) - same as zpos (?)
458 Contact: Emil Velikov, respective driver maintainers
462 Use struct dma_buf_map throughout codebase
463 ------------------------------------------
465 Pointers to shared device memory are stored in struct dma_buf_map. Each
466 instance knows whether it refers to system or I/O memory. Most of the DRM-wide
467 interface have been converted to use struct dma_buf_map, but implementations
468 often still use raw pointers.
470 The task is to use struct dma_buf_map where it makes sense.
472 * Memory managers should use struct dma_buf_map for dma-buf-imported buffers.
473 * TTM might benefit from using struct dma_buf_map internally.
474 * Framebuffer copying and blitting helpers should operate on struct dma_buf_map.
484 Make panic handling work
485 ------------------------
487 This is a really varied tasks with lots of little bits and pieces:
489 * The panic path can't be tested currently, leading to constant breaking. The
490 main issue here is that panics can be triggered from hardirq contexts and
491 hence all panic related callback can run in hardirq context. It would be
492 awesome if we could test at least the fbdev helper code and driver code by
493 e.g. trigger calls through drm debugfs files. hardirq context could be
494 achieved by using an IPI to the local processor.
496 * There's a massive confusion of different panic handlers. DRM fbdev emulation
497 helpers have one, but on top of that the fbcon code itself also has one. We
498 need to make sure that they stop fighting over each another.
500 * ``drm_can_sleep()`` is a mess. It hides real bugs in normal operations and
501 isn't a full solution for panic paths. We need to make sure that it only
502 returns true if there's a panic going on for real, and fix up all the
505 * The panic handler must never sleep, which also means it can't ever
506 ``mutex_lock()``. Also it can't grab any other lock unconditionally, not
507 even spinlocks (because NMI and hardirq can panic too). We need to either
508 make sure to not call such paths, or trylock everything. Really tricky.
510 * For the above locking troubles reasons it's pretty much impossible to
511 attempt a synchronous modeset from panic handlers. The only thing we could
512 try to achive is an atomic ``set_base`` of the primary plane, and hope that
513 it shows up. Everything else probably needs to be delayed to some worker or
514 something else which happens later on. Otherwise it just kills the box
515 harder, prevent the panic from going out on e.g. netconsole.
517 * There's also proposal for a simplied DRM console instead of the full-blown
518 fbcon and DRM fbdev emulation. Any kind of panic handling tricks should
519 obviously work for both console, in case we ever get kmslog merged.
521 Contact: Daniel Vetter
525 Clean up the debugfs support
526 ----------------------------
528 There's a bunch of issues with it:
530 - The drm_info_list ->show() function doesn't even bother to cast to the drm
531 structure for you. This is lazy.
533 - We probably want to have some support for debugfs files on crtc/connectors and
534 maybe other kms objects directly in core. There's even drm_print support in
535 the funcs for these objects to dump kms state, so it's all there. And then the
536 ->show() functions should obviously give you a pointer to the right object.
538 - The drm_info_list stuff is centered on drm_minor instead of drm_device. For
539 anything we want to print drm_device (or maybe drm_file) is the right thing.
541 - The drm_driver->debugfs_init hooks we have is just an artifact of the old
542 midlayered load sequence. DRM debugfs should work more like sysfs, where you
543 can create properties/files for an object anytime you want, and the core
544 takes care of publishing/unpuplishing all the files at register/unregister
545 time. Drivers shouldn't need to worry about these technicalities, and fixing
546 this (together with the drm_minor->drm_device move) would allow us to remove
551 Contact: Daniel Vetter
555 Object lifetime fixes
556 ---------------------
558 There's two related issues here
560 - Cleanup up the various ->destroy callbacks, which often are all the same
563 - Lots of drivers erroneously allocate DRM modeset objects using devm_kzalloc,
564 which results in use-after free issues on driver unload. This can be serious
565 trouble even for drivers for hardware integrated on the SoC due to
566 EPROBE_DEFERRED backoff.
568 Both these problems can be solved by switching over to drmm_kzalloc(), and the
569 various convenience wrappers provided, e.g. drmm_crtc_alloc_with_planes(),
570 drmm_universal_plane_alloc(), ... and so on.
572 Contact: Daniel Vetter
576 Remove automatic page mapping from dma-buf importing
577 ----------------------------------------------------
579 When importing dma-bufs, the dma-buf and PRIME frameworks automatically map
580 imported pages into the importer's DMA area. drm_gem_prime_fd_to_handle() and
581 drm_gem_prime_handle_to_fd() require that importers call dma_buf_attach()
582 even if they never do actual device DMA, but only CPU access through
583 dma_buf_vmap(). This is a problem for USB devices, which do not support DMA
586 To fix the issue, automatic page mappings should be removed from the
587 buffer-sharing code. Fixing this is a bit more involved, since the import/export
588 cache is also tied to &drm_gem_object.import_attach. Meanwhile we paper over
589 this problem for USB devices by fishing out the USB host controller device, as
590 long as that supports DMA. Otherwise importing can still needlessly fail.
600 Enable trinity for DRM
601 ----------------------
603 And fix up the fallout. Should be really interesting ...
607 Make KMS tests in i-g-t generic
608 -------------------------------
610 The i915 driver team maintains an extensive testsuite for the i915 DRM driver,
611 including tons of testcases for corner-cases in the modesetting API. It would
612 be awesome if those tests (at least the ones not relying on Intel-specific GEM
613 features) could be made to run on any KMS driver.
615 Basic work to run i-g-t tests on non-i915 is done, what's now missing is mass-
616 converting things over. For modeset tests we also first need a bit of
617 infrastructure to use dumb buffers for untiled buffers, to be able to run all
618 the non-i915 specific modeset tests.
622 Extend virtual test driver (VKMS)
623 ---------------------------------
625 See the documentation of :ref:`VKMS <vkms>` for more details. This is an ideal
626 internship task, since it only requires a virtual machine and can be sized to
627 fit the available time.
631 Backlight Refactoring
632 ---------------------
634 Backlight drivers have a triple enable/disable state, which is a bit overkill.
637 1. Roll out backlight_enable() and backlight_disable() helpers everywhere. This
639 2. In all, only look at one of the three status bits set by the above helpers.
640 3. Remove the other two status bits.
642 Contact: Daniel Vetter
649 AMD DC Display Driver
650 ---------------------
652 AMD DC is the display driver for AMD devices starting with Vega. There has been
653 a bunch of progress cleaning it up but there's still plenty of work to be done.
655 See drivers/gpu/drm/amd/display/TODO for tasks.
657 Contact: Harry Wentland, Alex Deucher
662 There is support in place now for writing internal DRM clients making it
663 possible to pick up the bootsplash work that was rejected because it was written
666 - [v6,8/8] drm/client: Hack: Add bootsplash example
667 https://patchwork.freedesktop.org/patch/306579/
669 - [RFC PATCH v2 00/13] Kernel based bootsplash
672 Contact: Sam Ravnborg
679 Convert fbdev drivers to DRM
680 ----------------------------
682 There are plenty of fbdev drivers for older hardware. Some hardware has
683 become obsolete, but some still provides good(-enough) framebuffers. The
684 drivers that are still useful should be converted to DRM and afterwards
687 Very simple fbdev drivers can best be converted by starting with a new
688 DRM driver. Simple KMS helpers and SHMEM should be able to handle any
689 existing hardware. The new driver's call-back functions are filled from
692 More complex fbdev drivers can be refactored step-by-step into a DRM
693 driver with the help of the DRM fbconv helpers. [1] These helpers provide
694 the transition layer between the DRM core infrastructure and the fbdev
695 driver interface. Create a new DRM driver on top of the fbconv helpers,
696 copy over the fbdev driver, and hook it up to the DRM code. Examples for
697 several fbdev drivers are available at [1] and a tutorial of this process
698 available at [2]. The result is a primitive DRM driver that can run X11
701 - [1] https://gitlab.freedesktop.org/tzimmermann/linux/tree/fbconv
702 - [2] https://gitlab.freedesktop.org/tzimmermann/linux/blob/fbconv/drivers/gpu/drm/drm_fbconv_helper.c
projects / linux.git / blob