Do I need to disable the AMD plugin when using the cuda plugin?

[GYDmedwin]

Description
Hi,
I want to checkpoint a Resnet job trained on an A100 using CRIU and the cuda plugin，but AMD errors are always reported. Do I need to comment out AMD-related content during compilation?

CRIU version: 4.0 CUDA Driver : 560
No container used and cuda-checkpoint is normal.

THANKS！

CRIU logs and information:

(00.600145) plugin: `amdgpu_plugin' hook 2 -> 0x730131770335
(00.600161) Error (amdgpu_plugin.c:1203): amdgpu_plugin: fstat error for /dev/kfd: No such file or directory
(00.600170) ----------------------------------------
(00.600175) Error (criu/cr-dump.c:1681): Dump files (pid: 3515192) failed with -1
(00.600179) Waiting for 3515192 to trap
(00.600188) Daemon 3515192 exited trapping
(00.600192) Sent msg to daemon 3 0 0
pie: 3515192: __fetched msg: 3 0 0
pie: 3515192: 3515192: new_sp=0x7a4028df37c8 ip 0x7a42faa90fd7
(00.620620) 3515192 was trapped
(00.620643) 3515192 was trapped
(00.620645) 3515192 (native) is going to execute the syscall 15, required is 15
(00.620658) 3515192 was stopped
(00.620818) net: Unlock network
(00.620824) amdgpu_plugin: finished  amdgpu_plugin (AMDGPU/KFD)
(00.620894) cuda_plugin: finished cuda_plugin stage 0 err -1
(00.784858) cuda_plugin: resuming devices on pid 3515192
(00.784864) cuda_plugin: Restore thread pid 3515332 found for real pid 3515192
(00.784948) Unfreezing tasks into 1
(00.784950)     Unseizing 3515192 into 1
(00.785380) Error (criu/cr-dump.c:2111): Dumping FAILED.

CRIU full dump/restore logs:

(paste your output here)

Output of `criu --version`:

(paste your output here)

Output of `criu check --all`:

(paste your output here)

Additional environment details:

[rst0git]

Do I need to comment out AMD-related content during compilation?

By default, the AMD GPU plugin for CRIU is installed in /usr/lib/criu/amdgpu_plugin.so. You can safely remove this file to avoid loading the plugin during checkpoint/restore.