Re: CVS commit: src/sys/kern

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, Dec 05, 2018 at 10:02:44AM +0100, Maxime Villard wrote:
> Le 03/12/2018 à 19:35, Manuel Bouyer a écrit :
> > On Mon, Dec 03, 2018 at 12:54:26PM +0100, Maxime Villard wrote:
> > > In other words, 80% of KASLR is enabled by default, regardless of #ifdef
> > > KASLR. Therefore, it is wrong to add an ifdef, because in either case we
> > 
> > So there's no way to completely disable KASLR now ?
> > Although I admit it's usefull to have it on by default, there should be a way
> > to turn it off for low-level debugging
> 
> No, I thought about that but in the end there is no way, because I didn't
> want to introduce another ton of #ifdefs, there are already too many. In
> general, you don't actually need to turn it off for debugging, it's enabled
> very early, and there is no reason for things to go wrong before

There's no reason until you run into one :)
I'm also concerned if things go wrong *after*, but before ddb is useable.
This is where reproductible boots can be important (I've been there).

> (locore.S, which I already largely fixed). The only special case is the PTE
> space; but given that it is recursive, even when it's a static location, it
> is still very difficult to debug.

I know, I've been there :)

> 
> Now that the design is stable and simpler than before, it should be less
> complicated to add an option to turn it off; but I still wouldn't want to do
> that because it adds complexity for no good reason.

Maybe "turn it off" is not the right word, it's more "reproductible boots":
two consecutive boots gives identical and predictible memory contents
(up to some point, of course).

as I understand KASLR (but I didn't look at the code), addresses are choosen
at random. So I guess there's a random number generator in the process.
Maybe it would be as simple as making the random number generator return
a constant value ?

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--