Re: [PATCH] random: do not use jump labels before they are initialized

["Jason A. Donenfeld" <Jason@zx2c4.com>]

Hi Ard,

On Tue, Jun 07, 2022 at 02:03:28PM +0200, Ard Biesheuvel wrote:
> On Tue, 7 Jun 2022 at 13:35, Jason A. Donenfeld <Jason@zx2c4.com> wrote:
> >
> > Hi Ard,
> >
> > On Tue, Jun 07, 2022 at 01:10:52PM +0200, Ard Biesheuvel wrote:
> > > Fair enough. What I would like is to remove the need to play around
> > > with the placement of jump_label_init() across architectures. Jump
> > > labels are fundamentally a performance optimization, so unless you can
> > > explain how setting it as early as possible makes a material
> > > difference, performance or otherwise, I really think we should pursue
> > > a solution that does the static key manipulation at some later time.
> >
> > Alright. It sounds like Catalin also prefers the same. This seems simple
> > enough with minimal downsides: https://lore.kernel.org/lkml/20220607113238.769088-1-Jason@zx2c4.com/
> >
> 
> That looks simple enough. Do we risk causing any boot stalls due to
> the crediting being deferred? Or new warnings about randomness being
> used before CRNG is ready?

We don't risk boot stalls. But there will be warnings for developers who
have enabled the CONFIG_WARN_ALL_UNSEEDED_RANDOM debug option.


> > So maybe we should just go that route.
> >
> 
> It is not my preferred approach, but I can live with it.

I'm not sure what your preferred approach is at this point in time
actually. I'll summarize all the approaches discussed so far:

1) Fix archs to initialize jump labels earlier:
   https://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux.git/commit/?id=73e2d827a501
   https://lore.kernel.org/lkml/20220603121543.360283-1-Jason@zx2c4.com/

2) Defer mixing & crediting until random_init():
   https://lore.kernel.org/lkml/20220607111514.755009-1-Jason@zx2c4.com/

3) Defer crediting (but not mixing) until random_init():
   https://lore.kernel.org/lkml/20220607113238.769088-1-Jason@zx2c4.com/

4) Defer changing the static branch (but neither mixing nor crediting) until random_init():
   https://lore.kernel.org/lkml/20220607100210.683136-1-Jason@zx2c4.com/


My first choice is (1) if it's feasible.

(2) is not possible without introducing a copy, so that's out.

What's your preferred approach? Or is there a number 5 you have in mind?

Jason