[BunnyFl]

What

sceKernelLoadStartModule and sceKernelStopUnloadModule ignored their flags and opt arguments, and sceKernelStopUnloadModule treated handle 0 like any unknown handle. Firmware 13.42 refuses these with 0x80020016 (EINVAL) before it does anything else, and this change does the same:

• sceKernelLoadStartModule: nonzero flags or a non-null opt returns 0x80020016, and res is left alone.
• sceKernelStopUnloadModule: nonzero flags, a non-null opt or handle 0 returns 0x80020016, and res is left alone.
• sceKernelLoadStartModuleForSysmodule forwards flags as 0. The firmware export adds 0x10000 to flags and jumps past the public entry's flags check, so a sysmodule load with flags still loads. A non-null opt is refused on both paths.

Firmware 13.42 system/common/lib/libkernel_sys.sprx (sha256 3d361753f6c472cd5442bf92652b4a1dee04051297a0155fe197ede0c2002d4a), read from disassembly, not run on a console:

sceKernelLoadStartModule (wzvqT4UqKX8), 0x35c70
35c70: test %ecx,%ecx            ; flags
35c72: je   0x35c7a
35c74: mov  $0x80020016,%eax
35c7a: xor  %ecx,%ecx
35c7c: jmp  0x35c90              ; shared body

sceKernelLoadStartModuleForSysmodule (Gg3+yWL6DWU), 0x363e0
363e0: or   $0x10000,%ecx
363e6: jmp  0x35c90

shared body, 0x35c90
35ca4: mov  $0x80020016,%eax
35ca9: test %r8,%r8              ; opt
35cac: jne  0x35e0a              ; return

sceKernelStopUnloadModule (QKd0qM58Qes), 0x36400
36411: test %ecx,%ecx            ; flags
36415: mov  $0x80020016,%eax
3641d: test %r8,%r8              ; opt
36427: jne  0x367ea              ; return
3642d: test %ebx,%ebx            ; handle
3642f: je   0x364db              ; 0x80020016 at 0x36502

The debt file gains the gap this doesn't close. sceKernelStopUnloadModule never runs the module's module_stop and stores 0 in res. At 0x36657 firmware calls module_stop with args and argp and stores its result in res (0x36663). It skips that call while a count at +0x6c is above 1 (0x364b1–0x36557), which reads like a load count. It returns 0x80020010 on a path (0x3648d–0x36583) that appears to cover another thread still starting the module. The loader reads only a start entry and initializers from the guest lifecycle metadata (core/libs/prx/libkernel/Module/src/DynamicLoader.cpp:221 and :267), with no stop entry, so a follow-up needs the relinker to record one.

Tested

• Windows 11, WinLibs GCC, Release, head caca1e4031: guest_dynamic_loader, guest_kernel_sys_module_internal and the four guest_deferred_module_* tests pass (6 of 6).
• New rows in core/libs/tests/GuestDynamicLoader.cpp load the test system module with flags 1 and with an opt block (refused, res untouched), then load it, refuse to unload it with flags 1, with an opt block and as handle 0, call into it, and unload it. New rows in core/libs/tests/GuestKernelSysModuleInternal.cpp check that sceKernelLoadStartModuleForSysmodule with flags 0x10000 still reaches the missing-file error, and that an opt block is refused.
• Negative control: both tests built against main's core/libs/prx/libkernel/Module/src/Module.cpp and core/libs/prx/libkernel_sys/Export.cpp abort (0xc0000409, 2 of 2 failed; guest_dynamic_loader stops at GuestDynamicLoader.cpp:209, the flags row).
• guest_runtime_lifecycle and guest_dynamic_loader_fixed_base_tests are registered on Linux only, and I didn't run them. Both pass flags 0 and a null opt.
• tools/check_conventions.py --base upstream/main on the committed head caca1e4031: all checks pass.
  • Correction: the first head, 5ee657294b, had a debt-length warning (the libkernel_sys.md module info entry reached 1186 characters). I had run the check before committing, and it reads committed Git objects (CONTRIBUTING.md, "reads pull request Git objects"), so it didn't see the debt change. The second commit moves the sysmodule flags note into its own entry, which fixes it.
• tools/pr_overlap.py --branch: the code-file overlaps are fix(relinker): match AppContent utility import to provider #13486, feat(libkernel_sys): implement mount, unmount and nmount privilege checks #15037, fix(libkernel_sys): reject a null aio_suspend list and a negative kenv length #15115, feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021, fix(libkernel): provide default arguments block for DT_INIT on Linux #6149, feat(libkernel_sys): implement sceKernelFlock over flock #9097 and feat(libkernel_sys): implement sigqueue #9389. In their diffs only fix(libkernel): provide default arguments block for DT_INIT on Linux #6149 names one of these functions, in a debt line about Linux DT_INIT arguments, and none changes them. The many debt-file overlaps are other entries.
• No title was run.

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
• Fixes: none
• Unblocks: none
• GPUs tested: n/a
• AI-assisted: yes

🤖 Generated with Claude Code