Hello,
I always get the error message nvbuf_utils:
dmabuf_fd -1853180096 mapped entry NOT found
Caught SIGSEGV
When I try to free a std::vector<std::vector<NvBufSurface*>>called buffer in a loop over each element.
Something like this:
for(int i=0; i<x; i++){
for(int j=0; j<y; j++){
NvBufSurfaceDestroy(buffer.at(i).at(j))
}
}
x and y are the same for the creation of the NvBufSurface.
Any ideas on why this might be happening.
Thanks.
Hi,
The dmabuf fd looks invalid. Do you call NvBufSurfaceCreate() to create the buffer? Or how the buffer is created?
Hi @DaneLLL ,
Yes it is created using NvBufSurfaceCreate followed by NvBufSurfaceMemSet.
The NvBufSurfaceCreate is called with the following parameter:
params.gpuId = DEFAULT_GPU_ID;
params.memType = NVBUF_MEM_SURFACE_ARRAY;
params.isContiguous = false;
params.width = width;
params.height = height;
params.size = 0;
params.layout = NVBUF_LAYOUT_PITCH;
params.colorFormat = NVBUF_COLOR_FORMAT_NV12;
Because I allocate upt to 100 NvBufSurfaces I checked my Ram, but this seems to be fine. Is there an artifical limit on how much NvBufSurfaces you can have?
Hi,
If you can create the buffer successfully, should be able to destroy it. Could you print out buffer.at(i).at(j) for a check? To make sure the fd is valid before calling NvBufSurfaceDestroy().
Hi @DaneLLL
The output is:
Vector of vectors
ptr: 0
ptr: 5ec0bfa0
ptr: 0
ptr: 5ec0c180
ptr: 0
ptr: 5ec0c520
ptr: 0
ptr: 5ec0c1d0
ptr: 0
ptr: 5ec0c260
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
ptr: 0
ptr: 5ec0c2f0
But now I am wondering,
my function for allocating NvBufSurface is this:
void AllocateNvBufSurface(NvBufSurface **surface, int width, int height, NvBufSurfaceLayout layout) noexcept {
NvBufSurfaceCreateParams params;
if (*surface != nullptr) {
NvBufSurfaceDestroy(*surface);
*surface = nullptr;
}
params.gpuId = DEFAULT_GPU_ID;
params.memType = NVBUF_MEM_SURFACE_ARRAY;
params.isContiguous = false;
params.width = width;
params.height = height;
params.size = 0;
params.layout = layout;
params.colorFormat = NVBUF_COLOR_FORMAT_NV12;
g_print("ptr: %x\n", *surface);
CHECK_NVBUFSURFACE_STATUS(NvBufSurfaceCreate(surface, 1, ¶ms), "Error: Could not allocate internal buffer for NvBufSurface at Startup");
g_print("ptr: %x\n", *surface);
CHECK_NVBUFSURFACE_STATUS(NvBufSurfaceMemSet(*surface, 0, 0, 0), "Error: Can't set all values of NvBufSurface at startup to 0");
}
And I get
ptr: 0
ptr: 5ebe0c40
ptr: 0
ptr: 5ec0acc0
ptr: 0
ptr: 5ec0b300
ptr: 0
ptr: 5ec0b940
Even for the surfaces I allocate separately and deallocate with NvBufSurfaceDestroy and that works without a problem.
Hi,
Is it possible the vectors are corrupted? Not sure if it is an issue but 5ec0c2f0 is read out for multiple times.
Hi @DaneLLL ,
I wouldn’t know how. So far I am only allocating and deallocating it and I don’t use it for anything yet.
@DaneLLL ,
It seems that I found the error. Within the allocation loop I create a structure that contains the NvBufSurface and some other variables. I create this on stack and push it back to the vector. It seems I exceeded the stack size. I now remodeled it with the creation of these structures on the heap and it doesn’t fail anymore.
Thanks for helping again.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.