Hi,
Is there a way to pin memory provided by a kernel-mode driver through mmap?
I have tried cudaHostRegister but it seems to only work when using MAP_ANONYMOUS (this defeats the goal - it zeros the memory and does not share with the driver).
Using cudaHostRegisterIoMemory gives me “operation not supported”
Alternatively can I somehow allocate CUDA memory in the driver?
This is for the Jetson Orin Nano.
Thanks!