Segmentation fault (core dumped) happend in VkBlobAllocator::fastFree(VkImageMemory* ptr)

[arnoldfychen]

error log | 日志或报错信息 | ログ

Segmentation fault (core dumped)

I got the backtrace of the segmentation fault by gdb:

Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x0000007f748cbed8 in ?? () from /usr/lib/aarch64-linux-gnu/tegra/libnvidia-eglcore.so.32.5.0
[Current thread is 1 (Thread 0x7f79e21010 (LWP 30466))]
(gdb) bt
#0 0x0000007f748cbed8 in () at /usr/lib/aarch64-linux-gnu/tegra/libnvidia-eglcore.so.32.5.0
#1 0x0000007f748cc2d8 in () at /usr/lib/aarch64-linux-gnu/tegra/libnvidia-eglcore.so.32.5.0
#2 0x0000007f749dfb10 in () at /usr/lib/aarch64-linux-gnu/tegra/libnvidia-eglcore.so.32.5.0
#3 0x0000007f749e5fec in () at /usr/lib/aarch64-linux-gnu/tegra/libnvidia-eglcore.so.32.5.0
#4 0x0000005567aeeb8c in ncnn::VkBlobAllocator::fastFree(ncnn::VkImageMemory*) ()
#5 0x0000005567afe4c0 in ncnn::VulkanDevicePrivate::destroy_dummy_buffer_image() ()
#6 0x0000005567afe700 in ncnn::VulkanDevice::~VulkanDevice() ()
#7 0x0000005567afe9d8 in ncnn::destroy_gpu_instance() ()
#8 0x0000007f90ad9e54 in __run_exit_handlers (status=1, listp=0x7f90bf85a0 <__exit_funcs>, run_list_atexit=127, run_list_atexit@entry=true, run_dtors=run_dtors@entry=true) at exit.c:108
#9 0x0000007f90ad9f8c in __GI_exit (status=) at exit.c:139
#10 0x0000007f83385424 in gtk_init () at /usr/lib/aarch64-linux-gnu/libgtk-x11-2.0.so.0
#11 0x0000007f917119dc in cvInitSystem () at /usr/lib/aarch64-linux-gnu/libopencv_highgui.so.4.1
#12 0x0000007f91714dbc in cvNamedWindow () at /usr/lib/aarch64-linux-gnu/libopencv_highgui.so.4.1
#13 0x0000007f91715650 in cvShowImage () at /usr/lib/aarch64-linux-gnu/libopencv_highgui.so.4.1
#14 0x0000007f9170c900 in cv::imshow(std::__cxx11::basic_string<char, std::char_traits, std::allocator > const&, cv::_InputArray const&) ()
at /usr/lib/aarch64-linux-gnu/libopencv_highgui.so.4.1
#15 0x0000005567ac9fd4 in main ()

context | 编译/运行环境 | バックグラウンド

The latest ncnn + vulkan (1.2.141 or 1.2.193) + Jetpack4.5 on NVIDIA Jetson Nano

how to reproduce | 复现步骤 | 再現方法

Under my work directory /home/ubuntu/ :

1. git clone -b v1.2.141 https://github.com/KhronosGroup/Vulkan-Loader.git or
   git clone https://github.com/KhronosGroup/Vulkan-Loader.git
2. cd Vulkan-Loader && mkdir build && cd build
3. ../scripts/update_deps.py
4. cmake -DCMAKE_BUILD_TYPE=Release -DVULKAN_HEADERS_INSTALL_DIR=$(pwd)/Vulkan-Headers/build/install ..
5. make
6. cd Vulkan-Headers
7. ln -s ../loader lib
8. export VULKAN_SDK=/home/ubuntu/Vulkan-Loader/build/Vulkan-Headers
9. git clone [email protected]:Tencent/ncnn
10. cd ncnn
11. git submodule update --init
12. mkdir build && cd build
13. cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/jetson.toolchain.cmake -DCMAKE_BUILD_TYPE=Release -DNCNN_VULKAN=ON -DNCNN_BUILD_EXAMPLES=ON ..
14. make
15. cd examples
16. copy scrfd_500m-opt2.bin and scrfd_500m-opt2.param to this folder and cp a test image test.jpg to this folder
17. ./scrfd test.jpg
18. You will see coredump happens almost when program ends.

more | 其他 | その他

I had an investigation and found the segmentation fault happens at line #1045 in void VkBlobAllocator::fastFree(VkImageMemory* ptr) in allocator.cpp :

vkDestroyImageView(vkdev->vkdevice(), ptr->imageview, 0);

[arnoldfychen]

Please note on Jetson Nano the vulkan version provided in JetPack4.5 is 1.2.141, I also had a try to use the latest vulkan 1.2.193 .so library,which I compiled out on Nano, to replace the 1.2.141 .so library under /usr/lib/aarch64-linux-gnu/tegra/:

cp /home/ubuntu/Vulkan-Loader/build/loader/libvulkan.so.1.2.193 /usr/lib/aarch64-linux-gnu/tegra/
cd /usr/lib/aarch64-linux-gnu/tegra
rm libvulkan.so.1
ln -s libvulkan.so.1.2.193 libvulkan.so.1
cd ..
rm libvulkan.so.1.2
ln -s tegra/libvulkan.so.1.2.193 libvulkan.so.1.2

but this segmentation fault still happened.