1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
|
===============================================================================
TODOs
===============================================================================
1. Base this on drm-next - WIP
2. Cleanup commit history
3. WIP - Drop page flip helper and use DRM's version
4. DONE - Flatten all DC objects
* dc_stream/core_stream/stream should just be dc_stream
* Same for other DC objects
"Is there any major reason to keep all those abstractions?
Could you collapse everything into struct dc_stream?
I haven't looked recently but I didn't get the impression there was a
lot of design around what was public/protected, more whatever needed
to be used by someone else was in public."
~ Dave Airlie
5. DONE - Rename DC objects to align more with DRM
* dc_surface -> dc_plane_state
* dc_stream -> dc_stream_state
6. DONE - Per-plane and per-stream validation
7. WIP - Per-plane and per-stream commit
8. WIP - Split pipe_ctx into plane and stream resource structs
9. Attach plane and stream reources to state object instead of validate_context
10. Remove dc_edid_caps and drm_helpers_parse_edid_caps
* Use drm_display_info instead
* Remove DC's edid quirks and rely on DRM's quirks (add quirks if needed)
"Making sure you use the sink-specific helper libraries and kernel
subsystems, since there's really no good reason to have 2nd
implementation of those in the kernel. Looks likes that's done for mst
and edid parsing. There's still a bit a midlayer feeling to the edid
parsing side (e.g. dc_edid_caps and dm_helpers_parse_edid_caps, I
think it'd be much better if you convert that over to reading stuff
from drm_display_info and if needed, push stuff into the core). Also,
I can't come up with a good reason why DC needs all this (except to
reimplement half of our edid quirk table, which really isn't a good
idea). Might be good if you put this onto the list of things to fix
long-term, but imo not a blocker. Definitely make sure new stuff
doesn't slip in (i.e. if you start adding edid quirks to DC instead of
the drm core, refactoring to use the core edid stuff was pointless)."
~ Daniel Vetter
11. Remove existing i2c implementation from DC
"Similar story for i2c, it uses the kernel's i2c code now, but there's
still a full i2c implementation hidden beneath that in
display/dc/i2caux. Kinda not cool, but imo ok if you fix that
post-merging (perhaps by not including any of this in the linux DC
code in the upstream kernel, but as an aux module in your internal
codebase since there you probably need that, same applies to the edid
parsing DC still does. For both cases I assume that the minimal shim
you need on linux (bit banging and edid parsing isn't rocket since) is
a lot less than the glue code to interface with the dc-provided
abstraction."
~ Daniel Vetter
12. drm_modeset_lock in MST should no longer be needed in recent kernels
* Adopt appropriate locking scheme
13. get_modes and best_encoder callbacks look a bit funny. Can probably rip out
a few indirections, and consider removing entirely and using the
drm_atomic_helper_best_encoder default behaviour.
14. core/dc_debug.c, consider switching to the atomic state debug helpers and
moving all your driver state printing into the various atomic_print_state
callbacks. There's also plans to expose this stuff in a standard way across all
drivers, to make debugging userspace compositors easier across different hw.
15. Move DP/HDMI dual mode adaptors to drm_dp_dual_mode_helper.c.
16. Move to core SCDC helpers (I think those are new since initial DC review).
17. There's still a pretty massive layer cake around dp aux and DPCD handling,
with like 3 levels of abstraction and using your own structures instead of the
stuff in drm_dp_helper.h. drm_dp_helper.h isn't really great and already has 2
incompatible styles, just means more reasons not to add a third (or well third
one gets to do the cleanup refactor).
18. There's a pile of sink handling code, both for DP and HDMI where I didn't
immediately recognize the standard. I think long term it'd be best for the drm
subsystem if we try to move as much of that into helpers/core as possible, and
share it with drivers. But that's a very long term goal, and by far not just an
issue with DC - other drivers, especially around DP sink handling, are equally
guilty.
19. The DC logger is still a rather sore thing, but I know that the DRM_DEBUG
stuff just isn't up to the challenges either. We need to figure out something
that integrates better with DRM and linux debug printing, while not being
useless with filtering output. dynamic debug printing might be an option.
|