Improving idle behavior in tickless systems

There are essentially two reasons, why this should not be a big problem for this prediction code:

First, user generated input interrupts are very few. Most people do not generate hundreds of keystrokes per second. And especially the average is really low. Therefore these interrupts hardly make a difference.

Second, while the user is producing input, especially moving the mouse, the interrupts are not that random as suggested. To generate randomness from these interrupts, one uses the least significant bits of the timing, i.e. nano- and microseconds. For power saving, we only need a rough estimate of the length of the idle period. Guessing the first bit that is not zero should be more than enough for this.

For network traffic, this is the same. Either there are few interrupts, or there is an ongoing transmission, which produces interrupts at quite regular intervals (or no interrupts at all, because the kernel switches to polling).