Nightly 3DS build crashes on opening sdmc
Description
I upgraded retroarch on my new3ds to latest nightly build, and every cores cause arm11 panic if I open sdmc.
Expected behavior
Retroarch can browse and load content
Actual behavior
Any attempt to read feom sdmc crashs the Arm11 subsystem. Libretro cores that don't require content can run as usual.
Steps to reproduce the bug
- Do a fresh upgrade to nightly build
- open retroarch, load any core, and enter "Load Content" > "sdmc:/"
Bisect Results
This accurs after I upgraded retroarch and its cores to nightly from a quite old version, with everything under /retroarch purged.
Version/Commit
- RetroArch: 1.19.1, built at Oct 22 2024
Environment information
- OS: Nintendo New3DS with Luma3DS v10.2.1 B9s, cfw "Sys 11.15.0-47J"
- Compiler: (I'm not building by myself but if helpful,) "GCC (10.2.0) 32-bit"
crash_dump_00000001.dmp I dumped this when it crashed.
One more thing, browsing sdmc:/ from Settings > Directory > (any entry) won't crash the machine. Using playlist can also avoid crashing and load contents.
Could you try an older nightly for me? From before my commit 790deebe4210208af7aa4c6331d8ced13867b20e which was yesterday my local time.
I'd try it out later.
I'd try it out later.
Did this end up working? I have this issue, attempting a full re-install as mentioned in a similar issue didn't work, and I'm unable to download a nightly as far back as @pstef mentioned.
EDIT: Crashing can by bypassed by adding game to Playlists, as navigating the SD card while managing them (or within Directory settings) doesn't crash. Note that Manual Scan can add games that don't match the database.
I also have experiences this same issue. Thanks to everyone for explaining how to work around it!
Version 1.20.0 is affected. The same crash happens on my n2ds, with @TheGershon's workaround working.
We should close this as a duplicate of #17430 and #17515.
I can't reproduce it with the most recent nightly gambatte.cia. I was able to play Alleyway (unzipped) and Dr. Mario (zipped).
Scratch that, I was finally able to bisect it down to c4685e8c319149dee1458fea5be347c5d9ea8679
I'm not sure what's exactly happening here, maybe I botched the git-bisect or there's some interaction with stack operations and at least one bug that overflows it. In any case I expect https://github.com/libretro/RetroArch/pull/17700 to improve the situation quite a bit, we'll see if it gets accepted.
Try nightly again, after https://github.com/libretro/RetroArch/pull/17700 (fixes) https://github.com/libretro/RetroArch/pull/17722 (improves).