Kernel failure - H100 HGX on ubuntu 22.04 /kernel 6.8.0-40 + driver 560.35.0 + cuda 12.6

Everything works fine with cuda 12.4 + driver 550.35.

with 12.6 + 560, the fabric-manager works fine. but nvidia-smi can’t detect an gpu and triggered the error msgin dmesg:

[ 6858.336904] NVRM: confComputeConstructEngine_IMPL: CPU does not support confidential compute.
[ 6858.336909] NVRM: nvAssertFailedNoLog: Assertion failed: 0 @ conf_compute.c:131
[ 6858.336915] NVOC: __nvoc_objDelete: Child class Spdm not freed from parent class ConfidentialCompute.NVRM: osInitNvMapping: *** Cannot attach gpu
[ 6858.336930] NVRM: RmInitAdapter: osInitNvMapping failed, bailing out of RmInitAdapter
[ 6858.336956] NVRM: GPU 0000:dd:00.0: RmInitAdapter failed! (0x22:0x38:744)
[ 6858.338673] NVRM: GPU 0000:dd:00.0: rm_init_adapter failed, device minor number 7
[ 6858.373651] NVRM: confComputeConstructEngine_IMPL: CPU does not support confidential compute.
[ 6858.373655] NVRM: nvAssertFailedNoLog: Assertion failed: 0 @ conf_compute.c:131
[ 6858.373660] NVOC: __nvoc_objDelete: Child class Spdm not freed from parent class ConfidentialCompute.NVRM: osInitNvMapping: *** Cannot attach gpu