Re: x86/random: Speculation to the rescue

[Linus Torvalds]

On Sat, Sep 28, 2019 at 4:53 PM Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> But hey, here's a made-up patch. It basically does jitter entropy, but
> it uses a more complex load than the fibonacci LFSR folding: it calls
> "schedule()" in a loop, and it sets up a timer to fire.

Ok, I'm sure a lot of people will end up finding this distasteful, and
I'll admit to just waffling about it myself.

But I am supposed to close the merge window today, and honestly, I
want _something_ to happen about the getrandom() issue during the 5.4
merge cycle.

So I had a few choices

 - just ignore things and hope some consensus happens

 - start the movement to a new getrandom() interface and encourage
user space to say "yeah, I don't need _secure_ random numbers"

 - or just say "hey, a lot of people find jitter entropy reasonable,
so let's just try this".

And I went with that last choice. If it works, it makes the
getrandom() interface changes a non-issue.

I'm not saying my patch is going to be the last word on the issue. I'm
_personally_ ok with it and believe it's not crazy, and if it then
makes serious people go "Eww" and send some improvements to it, then
it has served its purpose.

But I've committed that patch and the revert of the ext4 revert to a
local branch, I'll do some basic testing of it (which honestly on my
machines are kind of pointless, since all of them support rdrand), but
assuming it passes the basic smoke tests - and I expect it to - I'll
merge it for rc1.

I also have my old readdir branch that I decided I want to merge due
to the (completely independent) discussion about filldir issues, so
I'll probably end up delaying rc1 until tomorrow, but just a heads up.
I don't want to leave this until "some time later in the -rc series",
although I will be _more_ than happy to have people send me fixes to
my somewhat simplistic patch.

That said, my patch may be simplistic, but I suspect using a loop with
a real load like schedule() and arming a timer is a lot more realistic
than some of the jitter entropy papers with their _very_ trivial
LFSR's and some made-up pointer chasing.

But yeah, I think improvements to it are also not unexpected or unreasonable ;)

              Linus