[TheAggie]

What

mmap rejected any flag beyond MAP_PRIVATE | MAP_ANON for anonymous mappings. Quake II (PPSA99007, the Yamagi-based homebrew port) adds MAP_ALIGNED_SUPER (0x1000000, FreeBSD's MAP_ALIGNED(1)) to its hunk allocations of 2 MiB and more.

With that flag, an anonymous private mapping is now placed on a 2 MiB superpage boundary. DoMapAnon takes the alignment, defaulting to the 16 KiB page as before, and passes it to the existing MapAligned. Other flags are still rejected.

Test (guest_memory): a MAP_ALIGNED_SUPER mapping of 2 MiB + one page is 2 MiB aligned, writable at both ends and tracked with its full size, and unmaps. Another unsupported flag combination (0x2001002) is still rejected.

Tested

Windows 11 Pro (26200), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release.

• The full suite (489 tests) passes on current main with this PR and four other small PRs of mine merged together (SCHED_OTHER priority range, MAP_ALIGNED_SUPER, the compatible pad vibration mode, empty-PT_TLS stack-guard reads, pthread_attr scope/getters). It was run from the generated CTestTestfile.cmake commands because antivirus blocks ctest.exe on this machine.
• tools/check_conventions.py --base origin/main passes on this branch alone.

Not tested on Linux or macOS.

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 in TechnicalDebt
• 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
• AI-assisted: yes

🤖 Generated with Claude Code

[klosixgaming-arch]

Clean, correct support for MAP_ALIGNED_SUPER (0x1000000) in the anonymous mmap path: the flag is accepted on top of MAP_PRIVATE|MAP_ANONYMOUS (any other bit still returns ENOTSUPPORTED/45), the 2 MiB superpage alignment is threaded through DoMapAnon → MapAligned, and the default stays PS5_PAGE_SIZE so existing callers are unaffected. The test verifies the returned pointer is actually 2 MiB-aligned, that the full superpage + page span is readable/writable as a single allocation, and that an unsupported flag combination is still rejected.

Written with AI assistance.

[Dextroid17]

Verified on Linux — and this one has a behavioural differential.

• Arm A (this PR): guest_memory_tests builds and passes, exit 0.
• Arm B (control): the three DirectMemory files back at main with the test kept, rebuilt the same way → the test aborts, exit 134 (its Require is std::abort()), i.e. on main the flag is still refused. So both the acceptance and the superpage placement come from this change.

Claim cross-check: DoMapAnon takes the alignment and defaults it to PS5_PAGE_SIZE, the anonymous path passes SuperpageSize (2 MiB) only when GuestAlignedSuper (0x1000000) is set, and the guard became (flags & ~GuestAlignedSuper) != (GuestPrivate | GuestAnonymous) — every other flag is still refused. The test covers both halves: a 2 MiB + one page mapping lands on a 2 MiB boundary, is writable at both ends and tracked at its full size, and 0x2001002 is still rejected.

Two notes:

• A grep for the title's MAP_ALIGNED_SUPER finds nothing in the tree, because the code calls it GuestAlignedSuper and SuperpageSize. Worth spelling both out in the body, since a future reader will search the FreeBSD name.
• Question: does the alignment still hold when the guest passes an address hint (*addr != nullptr)? MapAligned receives *addr, and the test maps with a null hint — if a hint can win over the superpage alignment, a caller checking its own alignment would see something else.

The title is named (Quake II PPSA99007).

Not verified: Windows, the author's platform.

Written with AI assistance.

[oneandonlydean]

Same as #1868: Quake II stops at system (#1859) and mkstemp (#1613) on main first. Please reopen when it reaches the mmap call.

[adriadam10]

Checked on Linux: build and ctest green (guest_memory passes); with the three non-test files reverted the new test fails, so it covers the change. Other flag bits are still rejected, and the default alignment is unchanged. Confirmed also by @Dextroid17's A/B run. Regarding the note about Quake II stopping earlier: the change is generic and correct on its own, so merging.