feat(libkernel): implement msync

[hikazey]

What

Adds msync to libkernel, next to mmap, munmap and mprotect in DirectMemory/PosixMemory.cpp. It follows FreeBSD 11's kern_msync and vm_map_sync over the guest's mappings as sceKernelVirtualQuery reports them:

• The address is rounded down to the page and the length rounded up, with the same unsigned arithmetic as kern_msync. A range whose end wraps is EINVAL.
• MS_ASYNC | MS_INVALIDATE is EINVAL. Other flag bits are not checked, as in FreeBSD.
• An unmapped start, or any unmapped page in the range, is ENOMEM. Length 0 checks the mapping that holds the address (vm_map_sync widens start == end to that entry). Reservations count as mapped, since FreeBSD has a map entry for them.
• Success writes nothing back: the guest's mmap only creates private anonymous mappings, so there are no file mappings to sync.
• MS_INVALIDATE on a mapped range throws. FreeBSD returns EBUSY there when the range holds mlocked pages, and locked pages are not tracked. The ENOMEM check comes first, as in vm_map_sync.

One limitation is recorded in TechnicalDebt with the rest: a page munmaped from inside a larger mapping still counts as mapped, because sceKernelVirtualQuery keeps reporting the whole mapping after a partial unmap (on main too).

EVO-PLAYER PPSA99039 v0.11.0 (sainsaji/EVO-PLAYER-PS5) imports it for a mapped-file benchmark (projects/evoplayer/src/evo_iobench.c, msync(p, len, MS_SYNC)). It's also in the absent lists of other titles in #1578, where it was claimed.

Tested

• Title. EVO-PLAYER PPSA99039 v0.11.0 (release ZIP sha256 687c0f0f…5349, eboot.bin and the bundled libc.prx unwrapped from their SELFs), relinked with --windows on Windows 11. On main 364a56b with feat(libSceNet): implement sceNetResolverStartNtoaMultipleRecordsEx #1838, feat(libkernel): implement getdents and getdirentries #1634, feat(videoout): implement sceVideoOutGetResolutionStatus #1714 and feat(libkernel): implement the sigset functions #1641 merged locally (its earlier stops):
  • without this change it stops at FAIL: unresolved ELF import tZY4+SZNFhA from libkernel.prx (msync);
  • with it, it gets past and stops at KiJEPEWRyUY (sigaction, feat(libkernel): deliver sceKernelRaiseException on Linux #1431).
• New guest_msync test (Windows and Linux): success for the whole range, MS_ASYNC, an unaligned address inside the mapping, a range crossing a page boundary, length 0, an unknown flag bit and a reserved range, with errno untouched on success; EINVAL for MS_ASYNC | MS_INVALIDATE and a wrapping range; ENOMEM after munmap for the start, a range, length 0 and MS_INVALIDATE; the MS_INVALIDATE throw on a mapped range.
• Windows 11 Pro (26200), MinGW-w64 GCC 15.2.0 (WinLibs r7), Release, configured like CI: guest_msync 5/5; full suite 482/482. Mutations, each with the test kept, all fail: the mapped check removed, the MS_ASYNC | MS_INVALIDATE check removed, the throw moved before the mapped check, and the wrap check removed.
• Linux (Ubuntu 24.04 container, GCC 14.2, Release): guest_msync 5/5; full suite 488/488; the mapped-check mutation fails.
• Rebased onto main ae91d97: guest_msync 3/3 on Windows; tools/check_conventions.py --base origin/main passes.
• Not measured on a console: the semantics come from FreeBSD's source, as the TechnicalDebt line says.

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

[github-actions]

Progress report (8d47cfb → 0bff6da)

➖ System libraries: 85.46% (+0.0%, +1 implemented functions)

✅ 1 implemented

Library	Name
libkernel	msync_nid_postfix

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
#1633	core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp L1-2, L17, L115
#1909	core/libs/prx/libkernel/DirectMemory/PosixMemory.cpp L17

[adriadam10]

Checked msync against FreeBSD 11 kern_msync/vm_map_sync (page rounding, wrap and MS_ASYNC | MS_INVALIDATE give EINVAL, unmapped range gives ENOMEM, MS_INVALIDATE on a mapped range throws).

The branch conflicted with main only by adjacency (mlock/munlock in PosixMemory.cpp, the test block in libkernel/CMakeLists.txt). I rebased it keeping both sides, no logic changes. Full build and ctest on the rebased head: 525/525, guest_msync passes, and without the change it doesn't link, so it does exercise the code.