Re: Possible error in dtrace

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

Le 25/07/2018 à 16:09, Kamil Rytarowski a écrit :
> On 25.07.2018 16:01, Maxime Villard wrote:
> > Le 25/07/2018 à 15:54, Siddharth Muralee a écrit :
> > > Hello,
> > >         While going through sys/arch/amd64/pmap.h, I came across
> > > KERN_BASE[1] and confused it for KERNBASE[2]. Looking deeper into the
> > > same I think someone made the same mistake in dtrace_isa.c{3](I might
> > > be wrong).
> > >       I compared how the same variable was used in amd64[4] and
> > > i386[5]. There seems to be a mismatch. I computed the values for
> > > KERN_BASE and KERNBASE for amd64 and
> > >
> > >    * KERN_BASE = 0x800000000000
> > >    * KERNBASE = 0xffffffff80000000
> > >
> > > I think this may be an error. However, the names are definitely a bit
> > > confusing and there is no documentation to suggest what they are for.
> >
> > They are not the same. KERNBASE is the base where .text etc are located.
> > KERN_BASE is where the kernel map begins (VM_MIN_KERNEL_ADDRESS).
> >
> > In fact, KERN_BASE is incorrect, there should be a VA_SIGN_NEG() around the
> > multiplication. I've always known this, but I don't know if it's really
> > harmful, so by default I've never fixed it.
>
> I propose to either rename or fix KERN_BASE according to VA_SIGN_NET().
>
> These names are confusing.

The confusion likely comes from i386. On i386, VM_MIN_KERNEL_ADDRESS and
KERNBASE are equal.

The only dtrace user I see is dtrace_toxic_ranges(). I'm not sure what's
going on down there, but I think we want to exclude all of userland. So
the call should probably just be:

	(*func)(0, VM_MAX_ADDRESS)

Then we remove KERN_BASE.