xref: /openbmc/linux/drivers/gpu/drm/msm/NOTES (revision 588b48ca)
1NOTES about msm drm/kms driver:
2
3In the current snapdragon SoC's, we have (at least) 3 different
4display controller blocks at play:
5 + MDP3 - ?? seems to be what is on geeksphone peak device
6 + MDP4 - S3 (APQ8060, touchpad), S4-pro (APQ8064, nexus4 & ifc6410)
7 + MDP5 - snapdragon 800
8
9(I don't have a completely clear picture on which display controller
10maps to which part #)
11
12Plus a handful of blocks around them for HDMI/DSI/etc output.
13
14And on gpu side of things:
15 + zero, one, or two 2d cores (z180)
16 + and either a2xx or a3xx 3d core.
17
18But, HDMI/DSI/etc blocks seem like they can be shared across multiple
19display controller blocks.  And I for sure don't want to have to deal
20with N different kms devices from xf86-video-freedreno.  Plus, it
21seems like we can do some clever tricks like use GPU to trigger
22pageflip after rendering completes (ie. have the kms/crtc code build
23up gpu cmdstream to update scanout and write FLUSH register after).
24
25So, the approach is one drm driver, with some modularity.  Different
26'struct msm_kms' implementations, depending on display controller.
27And one or more 'struct msm_gpu' for the various different gpu sub-
28modules.
29
30(Second part is not implemented yet.  So far this is just basic KMS
31driver, and not exposing any custom ioctls to userspace for now.)
32
33The kms module provides the plane, crtc, and encoder objects, and
34loads whatever connectors are appropriate.
35
36For MDP4, the mapping is:
37
38  plane   -> PIPE{RGBn,VGn}              \
39  crtc    -> OVLP{n} + DMA{P,S,E} (??)   |-> MDP "device"
40  encoder -> DTV/LCDC/DSI (within MDP4)  /
41  connector -> HDMI/DSI/etc              --> other device(s)
42
43Since the irq's that drm core mostly cares about are vblank/framedone,
44we'll let msm_mdp4_kms provide the irq install/uninstall/etc functions
45and treat the MDP4 block's irq as "the" irq.  Even though the connectors
46may have their own irqs which they install themselves.  For this reason
47the display controller is the "master" device.
48
49For MDP5, the mapping is:
50
51  plane   -> PIPE{RGBn,VIGn}             \
52  crtc    -> LM (layer mixer)            |-> MDP "device"
53  encoder -> INTF                        /
54  connector -> HDMI/DSI/eDP/etc          --> other device(s)
55
56Unlike MDP4, it appears we can get by with a single encoder, rather
57than needing a different implementation for DTV, DSI, etc.  (Ie. the
58register interface is same, just different bases.)
59
60Also unlike MDP4, with MDP5 all the IRQs for other blocks (HDMI, DSI,
61etc) are routed through MDP.
62
63And finally, MDP5 has this "Shared Memory Pool" (called "SMP"), from
64which blocks need to be allocated to the active pipes based on fetch
65stride.
66
67Each connector probably ends up being a separate device, just for the
68logistics of finding/mapping io region, irq, etc.  Idealy we would
69have a better way than just stashing the platform device in a global
70(ie. like DT super-node.. but I don't have any snapdragon hw yet that
71is using DT).
72
73Note that so far I've not been able to get any docs on the hw, and it
74seems that access to such docs would prevent me from working on the
75freedreno gallium driver.  So there may be some mistakes in register
76names (I had to invent a few, since no sufficient hint was given in
77the downstream android fbdev driver), bitfield sizes, etc.  My current
78state of understanding the registers is given in the envytools rnndb
79files at:
80
81  https://github.com/freedreno/envytools/tree/master/rnndb
82  (the mdp4/hdmi/dsi directories)
83
84These files are used both for a parser tool (in the same tree) to
85parse logged register reads/writes (both from downstream android fbdev
86driver, and this driver with register logging enabled), as well as to
87generate the register level headers.
88