Other processes increasing alloc_page_nodemask rapidly in nvidia_poll when setting cudaDeviceScheduleBlockingSync flag in one process

[melfreq]

Hi,
When I used cudaSetDeviceFlags(cudaDeviceScheduleBlockingSync) in one process(producer), the number of alloc_page_nodemask called from nvidia_poll in cuda_EvtHandler thread increased rapidly in other processes(consumers).
producer → cuda – ipc → consumer1
– ipc → consumer2
Is this normal?
I thought cudaSetDeviceFlags only impact only the same process, in this case, producer.