When rsp is outside the kernel stack, the SIGUSR1 handler considers the context "out of syscall" and reports the real register values to wineserver. rsp pointing into the syscall frame is used for iret returns. Commit 0a5f7a710365 moved this earlier in the return path, creating a window in the regular direct return path where rsp points neither to the kernel stack, nor to the expected user stack. When a thread is interrupted by SIGUSR1 within this window, this exposes a bogus stack pointer to GetThreadContext(), breaking the garbage collector in Unity applications. Fix the regression by first switching to the real user stack (needed so we don't break things by touching other registers), then, only if an iret path is coming up, switching to the syscall frame pointer. This fixes the regression introduced by 0a5f7a710365, since GetThreadContext() will now observe the proper stack pointer during regular syscall returns. Further work would be needed to solve other corner cases and ensure that the register context returned by GetThreadContext() behaves exactly as on Windows (with no kernel-mode context info ever returned), but that is a larger/more invasive change. Fixes: 0a5f7a710365 ("ntdll: Switch to the user stack before restoring the %fs register.") Wine-Bug: https://bugs.winehq.org/show_bug.cgi?id=59333 Signed-off-by: Hoshino Lina lina@lina.yt -- v10: gitlab: Run ntdll tests with win64 architecture. ntdll: Abort wow64 tests early if wow64 is unavailable ntdll/tests: Add a test for bogus SP/IP values from GetThreadContext(). ntdll: Avoid exposing non-user rsp in normal syscall return path https://gitlab.winehq.org/wine/wine/-/merge_requests/10232