> I tried this too just now, and it didn't seem to make any difference.
> my test machine loses about 10 microseconds per second both when that
> dd pipeline is running and when the machine is idle.

how about killing a large process which holds many swap slots?
it seems that, without the patch, my machine can lose a few hundred ms
when killing a 1G byte process.

anyway, it's clear that it's bad to spin with IPL_HIGH.

YAMAMOTO Takashi