Re: [PATCH v2] x86, mm: set NX across entire PMD at boot

[Thomas Gleixner <tglx@linutronix.de>]

On Fri, 14 Nov 2014, Kees Cook wrote:
> On Fri, Nov 14, 2014 at 6:29 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> > On Fri, Nov 14, 2014 at 5:29 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> >> On Fri, Nov 14, 2014 at 12:45 PM, Kees Cook <keescook@chromium.org> wrote:
> >>> v2:
> >>>  - added call to free_init_pages(), as suggested by tglx
> >
> >> something is wrong:
> >>
> >> [    7.842479] Freeing unused kernel memory: 3844K (ffffffff82e52000 -
> >> ffffffff83213000)
> >> [    7.843305] Write protecting the kernel read-only data: 28672k
> >
> > ....
> > should use attached one instead.
> >
> > 1. should use _brk_end instead of &end, as we only use partial of
> >    brk.
> > 2. [_brk_end, pm_end) page range is already converted. aka
> >    is not wasted.
> 
> Are you sure? For me, _brk_end isn't far enough:

_brk_end is guaranteed to be <= _end. But we really want to use
_brk_end here, because if we have the following situation:

_brk_start:     0x03ff0000
_brk_end:       0x03ffff00
_end:           0x04016000

So we have the following PMDs setup:

0x03e00000     pmd rw nx        <- covers the top of .bss
                                   and the start  of .brk
0x04000000     pmd rw nx        <- covers the end of .brk
                                   and some random unused

So if _brk_end is less than 0x04000000, then cleanup_highmem() has
zapped the extra PMD already. So we don't want to call set_nx() on
that.

If _brk_end is >= 0x04000000 then we cover that last pmd with the
set_nx call.

Completely non obvious as anything else in that area.

Thanks,

	tglx