Messages in this thread | | | Subject | Re: [PATCH 01/10] random: make 'add_interrupt_randomness()' do something sane | From | Matt Mackall <> | Date | Thu, 05 Jul 2012 16:39:27 -0500 |
| |
On Thu, 2012-07-05 at 11:52 -0700, Linus Torvalds wrote: > On Thu, Jul 5, 2012 at 11:47 AM, Matt Mackall <mpm@selenic.com> wrote: > > > > I think you should demand a minimum number of events > HZ to actually > > credit any valid entropy. > > There already is. It's 1.
> If we don't get a single non-timer interrupt, this code will never be called.
From my read, this code path gets called on timer interrupts too. Thus the number of events per HZ will be HZ at a minimum on systems that haven't gone tickless. If such systems a) don't have a higher-res time source and b) are halted or equivalent, such samples will be completely deterministic. So the threshold for crediting entropy should be HZ + 1.
But perhaps I've missed something.
(As I expressed in my last message, this is strictly a correctness issue and not a practical one. I've long held that the entropy counting model is bogus and should be abandoned.)
-- Mathematics is the supreme nostalgia of our time.
| |