Hi,
I encountered a GPU crash. The crash dump shows that the error is DMA_Pagefault.
This crash happens on a reading buffer operation, and the GPU virtual address accessing is 0x00000025d1036000
The buffer’s details:
How can a page fault error occur while reading a fully resident buffer with a valid address?
The reading address is within the buffer’s memory location range. The address should be valid.
And the buffer state is fully resident. If I’m correct, the buffer resides in the GPU memory.
Even if the memory page has been evicted to the system memory, when a page fault is raised, it should only need to spend more time to reload the page back to the GPU memory. It shouldn’t crash on the page fault error.
Can anyone explain the behaviour of the GPU memory management that caused this crash?
Thank you!