Re: CVS commit: src/sys/arch

[Maxime Villard <max%m00nbsd.net@localhost>]

Le 13/05/2016 13:18, Martin Husemann a écrit :
> On Fri, May 13, 2016 at 07:50:20AM +0200, Maxime Villard wrote:
> > It's all explained in the commit messages that came afterwards.
>
> Sorry, but it is still completely unclear to me what the overall
> benefit is.

I've explained it on tech-kern a hour ago, but my mails take on average
~8 hours to reach the netbsd mailing lists.

> An executable read-only .rodata should be ok

No, it's not ok.

> and only if you happen
> to have it cross a large page boundary it will cost memory.
>
> So what was all this churn needed for?

I had to split twice, because the first split is done at the x86 level,
and i386 too is affected by it. The second one is done at the amd64
level without touching i386.

I couldn't split everything at once, because I would have had to fix the
i386 asm at the same time.

As a result of all of my changes, it is no longer possible to put a
shellcode in a static variable of the kernel (even though it's still
possible in its heap via the direct map).

If you think having a RWX data segment is a nice idea, then I can't do
anything for you.