Re: [PATCH v3 1/2] x86/mm: Identify the end of the kernel area to be reserved

[Greg KH]

On Tue, Jul 23, 2019 at 03:05:14PM +0200, Greg KH wrote:
> On Mon, Jul 15, 2019 at 01:16:48PM -0700, H.J. Lu wrote:
> > On Mon, Jul 15, 2019 at 3:35 AM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> > >
> > > On Mon, 15 Jul 2019, Thomas Gleixner wrote:
> > > > On Mon, 15 Jul 2019, Mike Lothian wrote:
> > > > > That build failure is from the current tip of Linus's tree
> > > > > If the fix is in, then it hasn't fixed the issue
> > > >
> > > > The reverted commit caused a build fail with gold as well. Let me stare at
> > > > your issue.
> > >
> > > So with gold the build fails in the reloc tool complaining about that
> > > relocation:
> > >
> > >   Invalid absolute R_X86_64_32S relocation: __end_of_kernel_reserve
> > >
> > > The commit does:
> > >
> > > +extern char __end_of_kernel_reserve[];
> > > +
> > >
> > >  void __init setup_arch(char **cmdline_p)
> > >  {
> > > +       /*
> > > +        * Reserve the memory occupied by the kernel between _text and
> > > +        * __end_of_kernel_reserve symbols. Any kernel sections after the
> > > +        * __end_of_kernel_reserve symbol must be explicitly reserved with a
> > > +        * separate memblock_reserve() or they will be discarded.
> > > +        */
> > >         memblock_reserve(__pa_symbol(_text),
> > > -                        (unsigned long)__bss_stop - (unsigned long)_text);
> > > +                        (unsigned long)__end_of_kernel_reserve - (unsigned long)_text);
> > >
> > > So it replaces __bss_stop with __end_of_kernel_reserve here.
> > >
> > > --- a/arch/x86/kernel/vmlinux.lds.S
> > > +++ b/arch/x86/kernel/vmlinux.lds.S
> > > @@ -368,6 +368,14 @@ SECTIONS
> > >                 __bss_stop = .;
> > >         }
> > >
> > > +       /*
> > > +        * The memory occupied from _text to here, __end_of_kernel_reserve, is
> > > +        * automatically reserved in setup_arch(). Anything after here must be
> > > +        * explicitly reserved using memblock_reserve() or it will be discarded
> > > +        * and treated as available memory.
> > > +        */
> > > +       __end_of_kernel_reserve = .;
> > >
> > > And from the linker script __bss_stop and __end_of_kernel_reserve are
> > > exactly the same. From System.map (of a successful ld build):
> > >
> > > ffffffff82c00000 B __brk_base
> > > ffffffff82c00000 B __bss_stop
> > > ffffffff82c00000 B __end_bss_decrypted
> > > ffffffff82c00000 B __end_of_kernel_reserve
> > > ffffffff82c00000 B __start_bss_decrypted
> > > ffffffff82c00000 B __start_bss_decrypted_unused
> > >
> > > So how on earth can gold fail with that __end_of_kernel_reserve change?
> > >
> > > For some unknown reason it turns that relocation into an absolute
> > > one. That's clearly a gold bug^Wfeature and TBH, I'm more than concerned
> > > about that kind of behaviour.
> > >
> > > If we just revert that commit, then what do we achieve? We paper over the
> > > underlying problem, which is not really helping anything.
> > >
> > > Aside of that gold still fails to build the X32 VDSO and it does so for a
> > > very long time....
> > >
> > > Until we really understand what the problem is, this stays as is.
> > >
> > > @H.J.: Any insight on that?
> > >
> > 
> > Since building a workable kernel for different kernel configurations isn't a
> > requirement for gold, I don't recommend gold for kernel.
> 
> Um, it worked before this commit, and now it doesn't.  "Some" companies
> are using gold for linking the kernel today...

Oh nevermind, I see the rest of the thread where it's now not being
allowed, sorry for the noise.

greg k-h