Replaces the old vendored BDK with the one from hekate v6.5.5 (CTCaer's
2025/2026 refactors: sdmmc, SE, minerva v1.6, display, heap, return
value homogenisation). OmniNX additions re-applied on top:
bq24193_charger_connected() and jc_clear_input(); s_vprintf now ships
with the BDK.
Ported to the new API:
- sd_mount(), sdmmc_storage_* and ini_parse() now return 0 on success.
The installer checked ini_parse() != 1, which would have made it miss
manifest.ini and run every install as a clean install.
- SD helpers come from bdk/storage/sd.c (source/nx_sd.* removed),
ffsystem.c from the BDK, diskio.c/ffconf.h from hekate (ffconf keeps
the sd:/ volume prefix).
- minerva_init() needs the minerva storage; the installer still drops
to 800 MHz afterwards.
- Payload launch follows hekate (sd_file_read, hw_deinit); coreboot
path removed upstream.
- Stack moved to 0x96000000: the old 0x83100000 is inside the new heap.
- GUI/USB/compression parts of the BDK are not built.
Install: staging and sd:/ are on the same card, so pack folders are now
moved with f_rename (a missing target folder moves in one step, existing
ones are merged file by file). Copying is only a per-entry fallback.
The staging cleanup afterwards only removes empty folders.
Deletion: one f_stat per list entry, no full recursive count before
deleting; large folders are detected with a probe that stops after 51
entries. Progress lines count processed entries instead of needing an
extra pass for a total.
Six copy/delete walkers (fs.c + install.c) are replaced by fs_delete_tree,
fs_move_tree and fs_move_file sharing one path buffer and one copy buffer
per run; the 'alternative copy method' retry is gone. extras/hekate-ams-pro
is moved too. Backups keep copy semantics.
log_write was the only vsnprintf user and pulled in newlib's printf,
dtoa, malloc and soft-float code (~44 KB). The payload now fits the
RCM limit again (146800 -> 102801 bytes); the build fails if it grows
past it. debug.ini log=1 writes sd:/config/omninx/install.log.
niklascfw
Niklas080208