1 ===============================================================================
3 ===============================================================================
5 1. Base this on drm-next - WIP
8 2. Cleanup commit history
11 3. WIP - Drop page flip helper and use DRM's version
14 4. DONE - Flatten all DC objects
15 * dc_stream/core_stream/stream should just be dc_stream
16 * Same for other DC objects
18 "Is there any major reason to keep all those abstractions?
20 Could you collapse everything into struct dc_stream?
22 I haven't looked recently but I didn't get the impression there was a
23 lot of design around what was public/protected, more whatever needed
24 to be used by someone else was in public."
28 5. DONE - Rename DC objects to align more with DRM
29 * dc_surface -> dc_plane_state
30 * dc_stream -> dc_stream_state
33 6. DONE - Per-plane and per-stream validation
36 7. WIP - Per-plane and per-stream commit
39 8. WIP - Split pipe_ctx into plane and stream resource structs
42 9. Attach plane and stream reources to state object instead of validate_context
45 10. Remove dc_edid_caps and drm_helpers_parse_edid_caps
46 * Use drm_display_info instead
47 * Remove DC's edid quirks and rely on DRM's quirks (add quirks if needed)
49 "Making sure you use the sink-specific helper libraries and kernel
50 subsystems, since there's really no good reason to have 2nd
51 implementation of those in the kernel. Looks likes that's done for mst
52 and edid parsing. There's still a bit a midlayer feeling to the edid
53 parsing side (e.g. dc_edid_caps and dm_helpers_parse_edid_caps, I
54 think it'd be much better if you convert that over to reading stuff
55 from drm_display_info and if needed, push stuff into the core). Also,
56 I can't come up with a good reason why DC needs all this (except to
57 reimplement half of our edid quirk table, which really isn't a good
58 idea). Might be good if you put this onto the list of things to fix
59 long-term, but imo not a blocker. Definitely make sure new stuff
60 doesn't slip in (i.e. if you start adding edid quirks to DC instead of
61 the drm core, refactoring to use the core edid stuff was pointless)."
65 11. Remove dc/i2caux. This folder can be somewhat misleading. It's basically an
66 overy complicated HW programming function for sendind and receiving i2c/aux
67 commands. We can greatly simplify that and move it into dc/dceXYZ like other
70 12. drm_modeset_lock in MST should no longer be needed in recent kernels
71 * Adopt appropriate locking scheme
73 13. get_modes and best_encoder callbacks look a bit funny. Can probably rip out
74 a few indirections, and consider removing entirely and using the
75 drm_atomic_helper_best_encoder default behaviour.
77 14. core/dc_debug.c, consider switching to the atomic state debug helpers and
78 moving all your driver state printing into the various atomic_print_state
79 callbacks. There's also plans to expose this stuff in a standard way across all
80 drivers, to make debugging userspace compositors easier across different hw.
82 15. Move DP/HDMI dual mode adaptors to drm_dp_dual_mode_helper.c. See
83 dal_ddc_service_i2c_query_dp_dual_mode_adaptor.
85 16. Move to core SCDC helpers (I think those are new since initial DC review).
87 17. There's still a pretty massive layer cake around dp aux and DPCD handling,
88 with like 3 levels of abstraction and using your own structures instead of the
89 stuff in drm_dp_helper.h. drm_dp_helper.h isn't really great and already has 2
90 incompatible styles, just means more reasons not to add a third (or well third
91 one gets to do the cleanup refactor).
93 18. There's a pile of sink handling code, both for DP and HDMI where I didn't
94 immediately recognize the standard. I think long term it'd be best for the drm
95 subsystem if we try to move as much of that into helpers/core as possible, and
96 share it with drivers. But that's a very long term goal, and by far not just an
97 issue with DC - other drivers, especially around DP sink handling, are equally
100 19. DONE - The DC logger is still a rather sore thing, but I know that the
101 DRM_DEBUG stuff just isn't up to the challenges either. We need to figure out
102 something that integrates better with DRM and linux debug printing, while not
103 being useless with filtering output. dynamic debug printing might be an option.
105 20. Use kernel i2c device to program HDMI retimer. Some boards have an HDMI
106 retimer that we need to program to pass PHY compliance. Currently that's
107 bypassing the i2c device and goes directly to HW. This should be changed.
109 21. Remove vector.c from dc/basics. It's used in DDC code which can probably
110 be simplified enough to no longer need a vector implementation.