[Bug 219216] sched_bind() blocks if the entropy pool is starved

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=219216

Fabian Keil <fk at fabiankeil.de> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |fk at fabiankeil.de

--- Comment #9 from Fabian Keil <fk at fabiankeil.de> ---
I can't reproduce the issue either. I'm using a kernel based on r318145:

CPU: Intel(R) Core(TM) i5-2520M CPU @ 2.50GHz (2491.96-MHz K8-class CPU)

I let the script run through 1680 cycles without noticeable issues.

While dd keeps a core busy as expected, rand_harvestq stays below 3%
cpu use according to top. I use vanilla powerd(8) which changes the
cpu frequency between 800 and 2501.

The fact that rand_harvestq is busy doesn't necessarily indicate that
the entropy pool is starved.

You could try running:
sudo dtrace -n 'fbt:kernel:random_harvest_*:entry {@[probefunc, stack()] =
count(); } tick-60sec {exit(0)}'
to see which random_harvest_* functions are called, from where and how often.

Additionally you could experiment with modifying the entropy sources with:
kern.random.harvest.mask
to see if it makes a difference.

Newly harvested entropy is mixed into the pool so adding more shouldn't
lower the "entropy quality" in the pool. Therefore it's not obvious to me how
the "poisoning" you mentioned would work.

-- 
You are receiving this mail because:
You are the assignee for the bug.