- previously, we left brightness voltage and LCD panel voltage on even though we disabled PWM and output signal, causing GSP to misdetect screen state in turn. Fix that (this fixes the issue mentioned above)
- Initialize the LCD controller over I2C, this fixes some rare sync issues some screen have with other Arm9 homebrew
This may not fix 100% of the issues, only the most egregious/dangerous
ones. There's still a lot of tech debt in Luma's baremetal code, that
necessitate a full rewrite to fix (not planned for this year)
Closes#2137.
Even though (when running TWL/AGB FIRM) the SoC is in O3DS mode, and the GPU also is,
as well as most other components behaving as such (external RAM, L2C not usable, etc.),
this is NOT the case for the LCD and adaptive backlight logic which retains FULL N3DS
functionality, including a feature where the window is blended with a given color depending
on the overall relative luminance of that window.
However, Nintendo's own code mistakenly assumes the opposite, and clearly so ("if GPU in N3DS mode"
checks, not passing N3DS extra adaptive backlight (ABL) to TWL/AGB_FIRM). This has implications:
- Powersaving (ABL) settings in TWL/AGB_FIRM is inconsistent with *both* O3DS (because the new RGB blend
LUT has been set to its current value by NATIVE_FIRM) and N3DS (because "pwn_cnt" and "inertia" are missing
their N3DS-only bits)
- "rave party" when booting into TWL/AGB_FIRM or O3DS NATIVE_FIRM without these regs (well, the LUT)
initialized. Easiest way to do so is by leveraging the "DSi autooboot" feature Luma provides. It is worth
noting at least the LUT survives hardware reboots (if Nintendo were using DSi software that was using
TLNC-based reboots, they wouldn't have noticed).
Only touch the autoboot path, for now
- Stop spamming mcu reads... use the GPIO pin to know if there's a MCU interrupt, instead
- Wait for the MCU interrupts to be raised when powering up LCD+backlights
- Also turn off LCD+backlight when needed
