[p3n1cillin]

What

The Dolphin core that RetroArch (PPSA99169) ships, cores/dolphin_libretro.so, stops on Windows a few seconds into a GameCube boot:

terminate called after throwing an instance of 'std::runtime_error'
  what():  sysctl_nid_postfix: unsupported MIB 6.12

MIB {6, 12} is {CTL_HW, HW_REALMEM}. Dolphin's MemPhysical reads it on FreeBSD as the machine's memory size, into a size_t, without checking the result.

sysctl now answers {6, 12}, and sysctlbyname("hw.realmem") maps to it. It returns the same value and checks as hw.physmem {6, 5}: 8 bytes holding the _SC_PHYS_PAGES pages of 16 KiB, read-only (EPERM for a write) and ENOMEM for a short buffer, the same handling FreeBSD's sysctl_hw_realmem gives it next to sysctl_hw_physmem. FreeBSD reports there the memory the firmware describes, which is at least hw.physmem; AnyPS5 has one memory size and the console's values are unknown. This is recorded under Unknown function info in docs/dev/debt/libkernel.md.

Tested

• RetroArch PPSA99169 v1.0.0-beta.1 on Windows 11 (MinGW GCC 15.2.0), converted with --executable picker/picker.bin --executable es-de/es-de.bin and started in its game mode (--ps5-mode=game). The game request names /app0/cores/dolphin_libretro.so and a 288-byte GameCube DOL I wrote: one text section at 0x80003100 holding b .. No game files were involved.
  • Before, it stops as above after about 7 s.
  • After, Dolphin logs Starting core = GameCube mode, loads its Vulkan shader caches and Booting from executable: /app0/selftest/loop.dol, then runs for the whole 75 s run.
  • The run also had my open feat(libkernel): select on pipes on Windows #5013 (select on pipes, which RetroArch's web server needs) merged.
• guest_system_configuration: the hw.physmem check now also runs for {6, 12} and hw.realmem (length query, value, short buffer, EPERM for a write through sysctl and sysctlbyname). All 124 libkernel tests pass on Windows; Linux was not built locally.
• python tools/check_conventions.py --base origin/main passes.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none (reaching RetroArch's cores needs feat(libkernel): select on pipes on Windows #5013)
• Unblocks: the Dolphin core of RetroArch (PPSA99169)
• AI-assisted: yes

🤖 Generated with Claude Code

[p3n1cillin]

Linux run of this change: Ubuntu in WSL2, GCC 15.2.0, Ninja Release, BUILD_TESTING=ON, full build. Built from a local integration head a14a49d8d, which is main 5d50886ee with this PR and my other open library PRs merged (#5013, #5009, #5061, #5078, #5098; no conflicts). All 121 libkernel tests and all 68 libc tests pass, including guest_system_configuration, guest_getrlimit and guest_select. Not run alone on Linux; no title run on Linux.

[coltken95]

Tested on Linux x86-64 (Release build), with b16f8f3 cherry-picked onto main 14036cb:

• With the PR: guest_system_configuration passes.
• With main's SystemConfiguration.cpp and the PR's test it aborts with
  sysctl_nid_postfix: unsupported MIB 6.12

It also cherry-picks cleanly together with #5078, which edits the same files.

[oneandonlydean]

Rebased onto main to resolve the conflict in docs/dev/debt/libkernel.md after #5185; no other change.