Not able to do pinning of cuda on 18.04 LTS on X86 PC

[apexintern1]

When I ran the command ./rdma-cuda, I am getting the following error ioctl(PIN_CUDA src) failed: ret=-1 errno=22

Can anyone let me know on how to resolve this issue? Thanks in advance.

[pyt-hnu]

I am facing the same problem as you, have you solved it?

[apexintern1]

No, I have not solved it. Do you have any ideas?

[pyt-hnu]

No, I have not solved it. Do you have any ideas?

I solved the problem by accident. I also test this program now(./rdma-cuda). But i stuck on this one :(ret = ioctl(fd, PICOEVB_IOC_H2C2H_DMA, &dma_params)). Corresponding to the kernel program in pevb_dma.

[apexintern1]

How did you solve it, did you change anything ./rdma-cuda cuda code? Because when I removed the if-else condition while allocating memory, make it allocate for DGPU using cudaMalloc(), I didn't this error "ioctl(PIN_CUDA src) failed: ret=-1 errno=22", but I got error while doing dma transfer, just like you got "ret = ioctl(fd, PICOEVB_IOC_H2C2H_DMA, &dma_params)"

[pyt-hnu]

I haven't change anything. But i add some "printf" functions to debug. and i haven't remove the if-else condititon while allocating memory. last, i meet same problem "ret = ioctl(fd, PICOEVB_IOC_H2C2H_DMA, &dma_params)", I figured out the problem.( static int pevb_dma(struct pevb *pevb, bool c2h)),but i'm still checking it out. I think the problem is probably the fpga, because i using a different FPGA.

[apexintern1]

No I think, if we program the FPGA according to the given instructions, then it should work fine. I am using the FPGA they have specified, still I am getting errors.

[pyt-hnu]

Thanks for your sharing, I think i still need to learn something about fpga and dma. As for the correctness of the program(rdma-cuda), I am a little skeptical about it.

[apexintern1]

Can you tell me what are all the changes you have done to solve the errors, and tell me exactly where did you change it.

[pyt-hnu]

Can you tell me what are all the changes you have done to solve the errors, and tell me exactly where did you change it.

Is your kernel loaded onto the FPGA? I'm not doing anything special,

[apexintern1]

Yes, kernel is loaded in my system. It also in-use for the FPGA.

[pyt-hnu]

Have you finished now?

[pateldipen1984-nv]

share the steps you used to compile the applications and picoevb kernel module. does it have cuda installed? How pico fpga and dGPU connected in your x86 system?

[xlyzzz]

I haven't change anything. But i add some "printf" functions to debug. and i haven't remove the if-else condititon while allocating memory. last, i meet same problem "ret = ioctl(fd, PICOEVB_IOC_H2C2H_DMA, &dma_params)", I figured out the problem.( static int pevb_dma(struct pevb *pevb, bool c2h)),but i'm still checking it out. I think the problem is probably the fpga, because i using a different FPGA.

hi, have you changed the fpga board to solve this problem? or use ( static int pevb_dma(struct pevb *pevb, bool c2h), how to use pevb_dma? thanke you