I am trying debug jetson-rdma-picoevb(rel-36+) on jetson orin nx.The FPGA is connected to NX with PCIe(0007:01:00.0). After executing rdma-cuda-c2h-perf, I saved the buf from the cudaHostAlloc . A total of 64k data is transferred, but some data is not transferred each time. The attachment is the data I saved, for example, the offset position is 0x2020 there is 2K data that has not been updated; The offset position is 0x4820 there are 32 bytes of data that have not been updated. I suspect that I need to perform a flush operation for the cache after the PCIe transfer is complete?What should I do if I need to?
output_binary_file0.zip (290 Bytes)
Hi,
Could you give the flush a try to see if it fixes the issue?
Thanks.
I’ve tried two flush cache methods, but to no avail.
1，flush_cache_all(); The results have not changed, and some of the address data has not been updated.
2，dma_sync_single_for_cpu(pevb->dev, dst->dmas[0].addr, dst->dmas[0].len, DMA_FROM_DEVICE); After the API is called, the system suspends and restarts.
Problem solved, FPGA transfer problem, sorry for wasting your time.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.