segfault using full screen vulkan

[agiera]

program aborts ~1sec after entering full screen while using vulkan backend
running pop os 20.04 with nvidia-450 vulkan 1.2.148 on a gtx 970m

stack trace:

Thread 1 (Thread 0x7fccf37fe700 (LWP 11693)):
#0  __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
        set = {__val = {1024, 140518235290912, 2505628, 1145044993, 10921804596590213251, 10700172839009257639, 3, 261993005056, 4294967310, 94614839236704, 0, 0, 0, 0, 0, 4458766951310848768}}
        pid = <optimized out>
        tid = <optimized out>
#1  0x00007fcd0280b859 in __GI_abort () at abort.c:79
        save_stage = 1
        act = {__sigaction_handler = {sa_handler = 0x7fccf37fbb80, sa_sigaction = 0x7fccf37fbb80}, sa_mask = {__val = {94562026071808, 140518235290496, 94562005406728, 140517923635440, 94562026071808, 140518235290528, 94562006011261, 94562026071808, 140518235290560, 140518235290576, 94562007941462, 140518235290608, 25483701028, 94562026071808, 4458766951310848768, 140518235290608}}, sa_flags = -288695431, sa_restorer = 0x7fccd4c7d000}
        sigs = {__val = {32, 0 <repeats 15 times>}}
#2  0x00005600ef276f72 in wxFatalSignalHandler(int) () at /home/agiera/dolphins/ishiiruka/Externals/wxWidgets3/src/unix/utilsunx.cpp:1399
#3  0x00005600eed3dcea in EMM::sigsegv_handler(int, siginfo_t*, void*) (sig=11, info=0x7fccf37fbdb0, raw_context=0x7fccf37fbc80) at /home/agiera/dolphins/ishiiruka/Source/Core/Core/MemTools.cpp:301
        old_sa = 0x5600efebc9a0 <EMM::old_sa_segv>
        context = 0x7fccf37fbc80
        sicode = 1
        bad_address = 0
        ctx = 0x7fccf37fbca8
#4  0x00007fcd04dc13c0 in <signal handler called> () at /lib/x86_64-linux-gnu/libpthread.so.0
#5  0x00007fccbddff929 in  () at /lib/x86_64-linux-gnu/libnvidia-glcore.so.450.66
#6  0x00007fccbde005e9 in  () at /lib/x86_64-linux-gnu/libnvidia-glcore.so.450.66
#7  0x00007fccbdea10a7 in  () at /lib/x86_64-linux-gnu/libnvidia-glcore.so.450.66
#8  0x00005600ef5ebdb9 in Vulkan::UtilityShaderDraw::BindPipeline() (this=0x7fccf37fc6d0) at /home/agiera/dolphins/ishiiruka/Source/Core/VideoBackends/Vulkan/Util.cpp:766
        pipeline = 0x7fcce0ec40f0
#9  0x00005600ef5eaa52 in Vulkan::UtilityShaderDraw::Draw() (this=0x7fccf37fc6d0) at /home/agiera/dolphins/ishiiruka/Source/Core/VideoBackends/Vulkan/Util.cpp:503
#10 0x00005600ef5eb0cb in Vulkan::UtilityShaderDraw::DrawQuad(int, int, int, int, int, int, int, int, int, int, int, float) (this=0x7fccf37fc6d0, dst_x=304, dst_y=0, dst_width=1312, dst_height=1080, src_x=0, src_y=0, src_layer=0, src_width=640, src_height=480, src_full_width=640, src_full_height=528, z=0) at /home/agiera/dolphins/ishiiruka/Source/Core/VideoBackends/Vulkan/Util.cpp:556
        u0 = 0
        v0 = 0
        u1 = 1
        v1 = 0.909090936
        w = 0
        vertices = {{Position = {-1, 1, 0, 1}, TexCoord = {0, 0.909090936, 0, 0}, Color = 4294967295}, {Position = {1, 1, 0, 1}, TexCoord = {1, 0.909090936, 0, 0}, Color = 4294967295}, {Position = {-1, -1, 0, 1}, TexCoord = {0, 0, 0, 0}, Color = 4294967295}, {Position = {1, -1, 0, 1}, TexCoord = {1, 0, 0, 0}, Color = 4294967295}}
#11 0x00005600ef5fae82 in Vulkan::VulkanPostProcessingShader::Draw(PostProcessor*, TargetRectangle const&, TargetSize const&, unsigned long, TargetRectangle const&, TargetSize const&, unsigned long, unsigned long, int, float) (this=0x7fcce0c63b80, p=0x7fcce0c92a80, dst_rect=..., dst_size=..., dst_tex=140517922021984, src_rect=..., src_size=..., src_tex=140517921393248, src_depth_tex=0, src_layer=0, gamma=1) at /home/agiera/dolphins/ishiiruka/Source/Core/VideoBackends/Vulkan/PostProcessing.cpp:332

The full stack trace is attached here.

I tried to link the vulkan sdk to get more info, but wasn't able to manage it. Is there a guide for that?

Let me know if I can provide more info or if there's any methodology for debugging I could try. Thanks!

[agiera]

I was able to get the validation layer hooked up. I've attached the output
vulkan.log

VUID-vkCmdBindPipeline-commandBuffer-recording(ERROR / SPEC): msgNum: -365328357 - Validation Error: [ VUID-vkCmdBindPipeline-commandBuffer-recording ] Object 0: handle = 0x7f6c90da52b0, type = VK_OBJECT_TYPE_COMMAND_BUFFER; | MessageID = 0xea39881b | You must call vkBeginCommandBuffer() before this call to vkCmdBindPipeline(). The Vulkan spec states: commandBuffer must be in the recording state (https://vulkan.lunarg.com/doc/view/1.2.148.0/linux/1.2-extensions/vkspec.html#VUID-vkCmdBindPipeline-commandBuffer-recording)

So it seems the command buffers aren't activated when they should be. Not sure if more concurrency control is needed? Maybe merging mainline dolphin's CommandBufferManager would fix it?

I also found that this was the latest commit on CommandBufferManager.cpp and it says vulkan is broken because post processing is not finished yet. Am I wasting my time trying to debug this or is that statement outdated?
@Tinob