[WRAPPER] Fix SDL Vulkan bridge2 passing wrong fnc2 (regression from c7bc29329)

[runlevel5]

Summary

This fixes a regression introduced in commit c7bc293 ("[WRAPPER] Add/Fix SDL_Vulkan_GetVkGetInstanceProcAddr wrapped function from SDL2/SDL3 to use bridge2") that breaks Vulkan in any SDL2/SDL3 game using SDL_Vulkan_GetVkGetInstanceProcAddr.

The Bug

In both wrappedsdl2.c (line 843) and wrappedsdl3.c (line 56), AddCheckBridge2() is called with my->SDL_Vulkan_GetVkGetInstanceProcAddr as the fnc2 parameter. This is the SDL function that returns the native vkGetInstanceProcAddr pointer — not the pointer itself.

When my_vkGetInstanceProcAddr in wrappedvulkan.c later calls getBridgeFnc2(R_RIP) to retrieve the native proc address resolver, it gets SDL_Vulkan_GetVkGetInstanceProcAddr instead of the actual vkGetInstanceProcAddr. So when the wrapper does getprocaddr(instance, name), it actually calls SDL_Vulkan_GetVkGetInstanceProcAddr(instance, name), which ignores both arguments and just returns the vkGetInstanceProcAddr pointer again.

This pointer then gets wrapped as if it were the resolved Vulkan function (e.g. vkEnumerateInstanceVersion). When the game calls through the bridge thinking it's calling vkEnumerateInstanceVersion(&version), it actually calls vkGetInstanceProcAddr(&version) — interpreting a stack pointer as a VkInstance handle. This triggers the Vulkan loader's "Invalid instance [VUID-vkGetInstanceProcAddr-instance-parameter]" validation error followed by SIGSEGV/SIGABRT.

The Fix

Pass (void*)emu->context->vkprocaddress instead, which already holds the actual native vkGetInstanceProcAddr function pointer (set on the preceding lines via SDL_Vulkan_GetVkGetInstanceProcAddr()). Same fix applied to both SDL2 and SDL3 wrappers.

Testing

I tested this on PPC64LE (POWER9) with SuperTuxKart 1.5. Before the fix, the game crashes immediately on Vulkan init with validation errors and SIGABRT. After the fix, Vulkan initializes correctly — detects AMD Radeon RX 6600 XT via RADV NAVI23 (Vulkan 1.4.328), creates command loader threads, and the game runs without crash.

This bug affects all architectures, not just PPC64LE — any box64 user running an SDL2/SDL3 Vulkan game will hit this crash.

[ptitSeb]

Ah indeed, nice debugging.

But I have an issue with the fix. It should use the result of the SDL call, no matter the previous state of emu->context->vkprocaddress

something like:

    void* procaddr= my->SDL_Vulkan_GetVkGetInstanceProcAddr();
    if(!emu->context->vkprocaddress)
        emu->context->vkprocaddress = (vkprocaddess_t)procaddr;

    if(emu->context->vkprocaddress)
        return (void*)AddCheckBridge2(my_lib->w.bridge, pFEpp, my_vkGetInstanceProcAddr, (void*)procaddr, 0, "vkGetInstanceProcAddr");

[runlevel5]

Good point — updated. Now always calls SDL_Vulkan_GetVkGetInstanceProcAddr() and stores the result in a local procaddr, which is passed directly as fnc2. The emu->context->vkprocaddress cache is only populated if it wasn't already set.

[ptitSeb]

Ok thanks.

(note: all those (void*) transtyping are useless, it's already void*)

[runlevel5]

(note: all those (void*) transtyping are useless, it's already void*)

Let me clean up

[runlevel5]

Removed the redundant (void*) cast on the procaddr assignment — it's already void* from the SDL call. Kept the one on the AddCheckBridge2 return since that's a uintptr_t to void* conversion. Force-pushed.

[ptitSeb]

LGTM, Thanks!