Atmosphere 1.2.4

Hotfix de dernière minute pour notre Custom Firmware préféré. La version 1.2.3 sortie ce matin est émaillée de bugs qui peuvent affecter certains jeux. SciresM a donc corrigé le tir en conséquence. En cadeau bonus, le temps de chargement d’Animal Crossing (moddé) avec la mise à jour maléfique 2.0.0 (qui s’enflamme sur le nombre de fichiers…plus de 99000) a été largement revu à la baisse, il semblerait que des optimisations aient été trouvées ! Attention, les sigpatches ne sont pas encore à jour, mais devraient arriver dans quelques minutes… sigpatches disponibles !

↓ Nouveautés / Corrections (cliquer pour voir) ↓'

1.2.4 is Atmosphère’s sixtieth official release.

Please be sure to update fusee when upgrading to 1.2.4. fusee-primary no longer exists, and will not work any more.

With thanks to the @switchbrew team, Atmosphère 1.2.4 is bundled with hbl 2.4.1, and hbmenu 3.5.0.

The following was changed since the last release:

  • Changes were made to the way fs.mitm builds images when providing a layeredfs romfs.
    • Cache management (to avoid unnecessary rebuild) was revised, to add a grace period of ~500ms-1s between process closing romfs image and ams.mitm needing to rebuild if romfs is re-opened.
      • This makes our cache much more effective, previously we were re-building romfs several times.
    • RomFS image ownership was overhauled, with a new reference-counting implementation added (used to implement the above grace period).
      • Certain games (e.g. Puyo Puyo Tetris 2, probably others) were sensitive to this timing, and could use access patterns which would trigger creation of romfs image while previous romfs image was in the middle of destructor.
      • This could cause a fatal error, because the destructor for the old image could run simultaneously with building the new image.
    • This also provides a speedup versus the 1.2.3 code, with Animal Crossing now taking ~8 fewer seconds to get past the Nintendo Switch logo.
  • General system stability improvements to enhance the user’s experience.

And the following was changed in 1.2.3:

  • Because ams.TMA is taking longer to develop than expected, experimental support for Atmosphère’s gdbstub as a standalone is now available.
    • Please Note: The GDBstub is new and may have bugs/need work. If you find issues, please report them to SciresM#0524 — all help finding/fixing bugs is appreciated, here.
      • Generally speaking, if you would like to report information about fixes needed/discuss development of the gdbstub, join ReSwitched’s #dev-support channel.
    • To enable it, set atmosphere!enable_standalone_gdbstub = u8!0x1 in system_settings.ini.
      • The standalone also requires atmosphere!enable_htc = u8!0x0, but this should be the case for everyone since ams.TMA isn’t actually usable yet.
    • Once enabled, open the devkitPro provided-gdb (aarch64-none-elf-gdb for 64-bit or arm-none-eabi-gdb for 32-bit).
      • The standalone stub exposes itself on port 22225 — so the command to connect is target extended-remote <ip address>:22225.
      • Type info os processes to get a list of process IDs that can be attached to.
        • The stub should work on both system programs, games, and homebrew — but please note that debugging certain processes (like sockets) can cause hang due to the stub using them itself.
    • Software break-points, hardware break-points, hardware watch-points, and hardware single-step are all supported/implemented.
    • The following monitor commands are currently supported:
      • monitor get info: Get process info, address space layout, and information on modules.
      • monitor get mappings: Get all memory mappings.
      • monitor get mapping <addr>: Get the memory mapping for a specific address.
      • monitor wait application: Causes the stub to wait for an application to be launched. The next application will be started suspended.
        • User is expected to send attach <pid> after launching, which will cause attach-on-first-instruction. Failure to attach may cause system instability, this probably needs work.
  • Changes were made to the way fs.mitm builds images when providing a layeredfs romfs.
    • Animal Crossing’s 2.0.0 update contains >99000 files, and has tables so big that we ran out of memory even after the optimizations made in 0.10.5.
      • Previously, we used fixed-sized 0x40000 work buffers for file/directory tables and simultaneously built hash/content tables in one loop over files/directories.
      • We now iterate over the file/directory tables multiple times, first once to determine the hash table indices, then repeatedly to build hash tables, then once to build content tables.
      • We also now allow smaller-than-0x40000 work buffers, trying half-as-big buffers until allocation succeeds (or work buffer would be <0x4000, which is a safeguard against truly horrible performance).
    • There is a slight speed penalty to these changes, but it’s on the order of seconds for the worst case (Animal Crossing) and trivial for most games with reasonable tables.
    • If you encounter a game that exhausts ams.mitm’s memory (crashing it) when loading layeredfs mods, please contact SciresM#0524.
      • It’s really hard to imagine any game being worse than Animal Crossing, but if it happens again I will drop everything to fix it as usual.
  • creport now attempts to parse symbol tables if present.
    • If a game executable has a symbol for a given address, the function-relative-offset will now be printed after the module-relative-offset.
  • General system stability improvements to enhance the user’s experience.

For information on the featureset supported by 1.2, please see the official release notes