fix(libkernel): fail dlopen of an unrelinked module without the Windows Bad Image dialog by johandavidcm · Pull Request #9832 · boykopovar/AnyPS5

johandavidcm · 2026-10-22T16:38:18Z

What

On Windows, dlopen passes a module that was never relinked (a PS5 ELF or SELF that a title loads with sceKernelLoadStartModule) to LoadLibraryExW. Windows then raises its modal Bad Image hard-error dialog (0xC000012F), and the calling thread blocks until someone answers it. After that, the load fails with error 193 and sceKernelLoadStartModule returns SCE_KERNEL_ERROR_ENOENT, with nothing in the log.

Yakuza 0 Director's Cut (PPSA31327) hits this whenever these modules are not relinked:

at every boot, for /app0/data/crm/ps5/crm_module_ps5.prx (the SEGA account module);
in chapter 1, for /app0/data/module/ps5/yakuza_minigame_ps5_retail.prx.

Unattended runs stall on the dialog.

No dialog: LoadWithoutErrorDialog sets SEM_FAILCRITICALERRORS on the calling thread for the LoadLibraryExW call only, keeps the loader's error code, and restores the previous thread mode. PatchedLibraries.cpp uses the same flag for its whole process. The module's initialization (InitializeDeferredModule) runs after the mode is restored.
Same result: the load fails at once with the same error. The title gets null / ENOENT and continues, as it did after someone answered the dialog.
Reporting: every failed load writes the dlerror text and prints it on stderr, once per file, with a [module] prefix. This applies on Windows and on the other hosts.
- A file that exists but has no relinked <name>.guest.prx gets: dlopen: <path> was not relinked: no <name>.guest.prx was found (Windows loader error 193); relink its directory with the relinker's --module-dir <dir> option.
- Any other failure gets: dlopen: cannot load <path> (<reason>).
Docs: USAGE.md mentions the message beside its existing note on error 193.

Overlaps:

feat(libkernel): add fallback search paths and loaded handle lookup i… #9415 changes the same Windows lines of dlopen (fallback paths). Whichever PR merges second swaps its LoadLibraryExW calls for LoadWithoutErrorDialog and routes its final failure through LoadFailure.
The macOS PRs feat: macOS host support (Apple Silicon via Rosetta 2, Vulkan on MoltenVK) #8658 and Macos port #7201 add __APPLE__ blocks between the POSIX lines changed here, and feat(aot): consolidate native ARM64 compiler, runtime and consumers #5021 touches the CMake lines near the new TIMEOUT. These are line-level conflicts only.

Tested

Windows 11 Pro 26200, WinLibs GCC 15.2.0, Release, BUILD_TESTING=ON, on head dc84bec (main a6646f8).

guest_dynamic_loader gains a check. dlopen of an existing 4-byte ELF stub that has no .guest.prx is called twice:

with the process error mode reset to the default, because CTest starts tests with critical-error boxes off;
with the thread mode set to SEM_NOGPFAULTERRORBOX.

Each call must return null, with a dlerror that names the file and --module-dir, and the thread mode must be unchanged afterwards. The missing-module dlerror must not mention --module-dir.

With main's DynamicLoader.cpp, the test shows the Bad Image dialog and fails on the test's new TIMEOUT (30 s, the CI limit).
With this change it passes in 0.1 s, prints the was not relinked line once for the two calls, and passes --repeat until-fail:20.

Other tests that pass: guest_dynamic_loader_variant_names, guest_deferred_module_{reload,pinned,parallel,reentrant}, guest_module_info, guest_module_{dir,directories,args,identity,fini} and patched_libraries.

PPSA31327 01.015.000, relinked with only --module-dir data/module/ps5, launched with the default error mode, from boot to gameplay:

With libkernel from main 80565c5, the Bad Image dialog for crm_module_ps5.prx appears 6 s after start.
With the libraries built from the change on main 0def01b (head 4894d1c, the same diff), no dialog appears, and stderr has one [module] dlopen: ...crm_module_ps5.prx was not relinked ... line. The title reaches gameplay, at about 200,000 draws per 10 s.

Linux: DynamicLoader.cpp and GuestDynamicLoader.cpp compile with GCC 15.2.0 (-Wall -Wextra, no warnings) in WSL Ubuntu. The test was not run there. On Linux and macOS the change is only the reporting: LoadFailure instead of Error(::dlerror()).

Not tested: macOS.

tools/check_conventions.py --base origin/main passes.

Checklist

MarcZX10

I reviewed head cc888176de25999eb68d6fc01d0cc641774b58da. python3 tools/check_conventions.py --base upstream/main and git diff --check upstream/main...refs/review/9832 pass. I also ran GCC 12.2 C++20 syntax-only compilation of the changed DynamicLoader.cpp and GuestDynamicLoader.cpp translation units; both compile successfully on Linux.

I did not run the CTest binary or the Windows modal-dialog reproduction here. The PR description reports the focused Windows test, repeated runs, and title reproduction; the PR's hosted Linux/Windows build jobs are skipped. The current main is two commits ahead of this head; a merge-tree check is clean. Please update to current main before merge and ensure the Windows focused test runs on the final head. I found no blocking issue in the reviewed diff.

JJenkx · 2026-10-22T16:56:45Z

Independent Windows run, as data only. Machine: Windows 11 Enterprise 10.0.22621, Ryzen 9 5950X, WinLibs GCC 15.2.0 UCRT r7, Release, BUILD_TESTING=ON, Python 3.15. #9832, #9793 and #9768 were merged together onto main 11f976b, without conflicts, and each test was run 3 times. guest_dynamic_loader passes 3 of 3.

AI-assisted (Claude Code).

github-actions · 2026-10-22T16:57:31Z

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
PR 6586	`core/libs/prx/libkernel/Module/src/DynamicLoader.cpp` L196
PR 9415	`core/libs/prx/libkernel/Module/src/DynamicLoader.cpp` L182-184
PR 9488	`docs/user/USAGE.md` L177

Some comparisons could not be completed because a pull request conflicts with the base branch. Review these shared files manually before concluding that there is no overlap.

Pull request	Shared files not compared
PR 4563	`core/libs/prx/libkernel/CMakeLists.txt`
PR 5021	`core/libs/prx/libkernel/CMakeLists.txt` `core/libs/prx/libkernel/Module/src/DynamicLoader.cpp` `docs/user/USAGE.md`
PR 5925	`docs/user/USAGE.md`
PR 6044	`core/libs/prx/libkernel/CMakeLists.txt` `core/libs/tests/GuestDynamicLoader.cpp`
PR 6171	`docs/user/USAGE.md`
PR 6469	`core/libs/prx/libkernel/CMakeLists.txt`
PR 6544	`core/libs/tests/GuestDynamicLoader.cpp`
PR 6788	`docs/user/USAGE.md`
PR 6830	`core/libs/prx/libkernel/CMakeLists.txt`
PR 6924	`docs/user/USAGE.md`
PR 6936	`core/libs/prx/libkernel/CMakeLists.txt`
PR 7201	`core/libs/prx/libkernel/CMakeLists.txt` `core/libs/prx/libkernel/Module/src/DynamicLoader.cpp`
PR 7642	`docs/user/USAGE.md`
PR 7800	`docs/user/USAGE.md`
PR 7918	`docs/user/USAGE.md`
PR 8520	`core/libs/prx/libkernel/CMakeLists.txt`
PR 8658	`core/libs/prx/libkernel/CMakeLists.txt` `core/libs/prx/libkernel/Module/src/DynamicLoader.cpp` `core/libs/tests/GuestDynamicLoader.cpp`
PR 9187	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9284	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9293	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9297	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9302	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9408	`docs/user/USAGE.md`
PR 9508	`docs/user/USAGE.md`
PR 9561	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9566	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9575	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9669	`core/libs/prx/libkernel/CMakeLists.txt`
PR 9693	`core/libs/prx/libkernel/CMakeLists.txt`

johandavidcm · 2026-10-22T17:22:13Z

Rebased onto main 499b1cf; the head is now 50a14b1. On that head (Windows 11 Pro 26200, WinLibs GCC 15.2.0, Release, BUILD_TESTING=ON), these pass:

guest_dynamic_loader, with --repeat until-fail:20.
guest_dynamic_loader_variant_names, guest_deferred_module_{reload,pinned,parallel,reentrant}, guest_module_info, guest_module_{dir,directories,args,identity,fini} and patched_libraries.
tools/check_conventions.py --base origin/main.

MarcZX10 · 2026-10-22T17:30:23Z

Follow-up on the refreshed head 50a14b1577: I compiled the exact PR DynamicLoader.cpp and GuestDynamicLoader.cpp with GCC 12.2 C++20 -Wall -Wextra -Werror -fsyntax-only on Linux; both passed. The PR is mergeable, but its base commit is still 499b1cf9fc, 24 commits behind current main 6a6a35fdf5c2, and the reported Windows test used 751b2881a. CONTRIBUTING requires a current-main base; please rebase and rerun the focused Windows guest_dynamic_loader/title check on that final head before merge. I found no new code blocker.

MarcZX10 · 2026-10-22T17:57:44Z

Thanks for rebasing and reporting the 20-repeat Windows guest_dynamic_loader run plus the related module tests. I also ran Linux -fsyntax-only -Wall -Wextra -Werror on the exact DynamicLoader.cpp and GuestDynamicLoader.cpp; both pass. Since that update, current main has advanced again: the head is now 52 commits behind the current snapshot, and this diff overlaps the open fallback/macos loader work noted by the overlap report. Please refresh onto current main and rerun the focused Windows test on the final combined head before approval. This is a contribution-rule/reviewability request, not a merge request.

…ws Bad Image dialog On Windows, dlopen passed a module that was never relinked to LoadLibraryExW, which raised the modal Bad Image hard-error dialog and blocked the calling thread until someone answered it. The load now runs with SEM_FAILCRITICALERRORS set for that thread only, so it fails at once with the same result as before. Every failed load is reported once per file on stderr with a [module] prefix; a file that exists with no relinked .guest.prx names the relinker's --module-dir option.

johandavidcm · 2026-10-22T18:21:58Z

The previous push (4894d1c) never reached this PR, which still showed 50a14b1, during today's GitHub incident with pull requests. Rebased again, onto main a6646f8; the head is now dc84bec, 0 commits behind. On that head (Windows 11 Pro 26200, WinLibs GCC 15.2.0, Release, BUILD_TESTING=ON):

guest_dynamic_loader passes --repeat until-fail:20.
guest_dynamic_loader_variant_names, guest_deferred_module_{reload,pinned,parallel,reentrant}, guest_module_info, guest_module_{dir,directories,args,identity,fini} and patched_libraries pass.
tools/check_conventions.py --base origin/main passes.

On 4894d1c, which is the same diff on main 0def01b:

With that main's DynamicLoader.cpp, the test showed the Bad Image dialog and failed on its 30 s TIMEOUT.
PPSA31327, launched with the default error mode, showed no dialog, logged one [module] ... crm_module_ps5.prx was not relinked ... line and reached gameplay.

The Tested section of the description names this head.

On the overlaps with the loader work, #9415 (fallback paths) and #8658/#6586 (macOS and Linux POSIX branch), this PR depends on none of them; each pair only conflicts on lines. I asked about the order in #9415.

AI-assisted: yes

github-actions Bot added the unblocks title Names a title ID on the Unblocks line label Oct 22, 2026

MarcZX10 reviewed Oct 22, 2026

View reviewed changes

This was referenced Oct 22, 2026

fix(relinker): preserve Windows extended-path arguments #9793

Merged

fix(libc): fail freopen of an empty name with ENOENT on Windows #9768

Merged

johandavidcm force-pushed the fix/libkernel-dlopen-unrelinked branch from cc88817 to 50a14b1 Compare October 22, 2026 17:21

johandavidcm force-pushed the fix/libkernel-dlopen-unrelinked branch from cc88817 to 50a14b1 Compare October 22, 2026 17:22

johandavidcm force-pushed the fix/libkernel-dlopen-unrelinked branch from 4894d1c to dc84bec Compare October 22, 2026 18:21

johandavidcm force-pushed the fix/libkernel-dlopen-unrelinked branch from 50a14b1 to 4894d1c Compare October 22, 2026 18:22

oneandonlydean self-assigned this Oct 22, 2026

github-actions Bot added libkernel build docs labels Oct 23, 2026

oneandonlydean merged commit 5e9f4b8 into boykopovar:main Oct 23, 2026
12 checks passed

maximilianfv mentioned this pull request Oct 24, 2026

fix(libkernel): keep dlerror messages whole #10627

Merged

12 tasks done

Navigation Menu

Repository navigation

fix(libkernel): fail dlopen of an unrelinked module without the Windows Bad Image dialog - #9832

fix(libkernel): fail dlopen of an unrelinked module without the Windows Bad Image dialog#9832
oneandonlydean merged 1 commit into
boykopovar:mainfrom
johandavidcm:fix/libkernel-dlopen-unrelinked

johandavidcm commented Oct 22, 2026 •

edited

Loading

Uh oh!

MarcZX10 left a comment

Uh oh!

JJenkx commented Oct 22, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Oct 22, 2026 •

edited

Loading

Uh oh!

johandavidcm commented Oct 22, 2026

Uh oh!

MarcZX10 commented Oct 22, 2026

Uh oh!

MarcZX10 commented Oct 22, 2026

Uh oh!

johandavidcm commented Oct 22, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Conversation

johandavidcm commented Oct 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What

Tested

Checklist

Uh oh!

MarcZX10 left a comment

Choose a reason for hiding this comment

Uh oh!

JJenkx commented Oct 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 22, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

johandavidcm commented Oct 22, 2026

Uh oh!

MarcZX10 commented Oct 22, 2026

Uh oh!

MarcZX10 commented Oct 22, 2026

Uh oh!

johandavidcm commented Oct 22, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

johandavidcm commented Oct 22, 2026 •

edited

Loading

JJenkx commented Oct 22, 2026 •

edited

Loading

github-actions Bot commented Oct 22, 2026 •

edited

Loading