On Fri, 22 Jun 2018, Kirill A. Shutemov wrote:
> On Fri, Jun 22, 2018 at 03:35:18PM +0000, Thomas Gleixner wrote:
> > > > Second thoughts.
> > > >
> > > > The only place where __pgtable_l5_enabled() is used in common.c is in
> > > > early_identify_cpu() which is marked __init. So how is that section
> > > > mismatch triggered?
> > >
> > > Yeah, it's not obvious:
> > >
> > > cpu_init()
> > > load_mm_ldt()
> > > ldt_slot_va()
> > > LDT_BASE_ADDR
> > > LDT_PGD_ENTRY
> > > pgtable_l5_enabled()
> >
> > How is that supposed to work correctly?
> >
> > start_kernel()
> > ....
> > trap_init()
> > cpu_init()
> >
> > ....
> > check_bugs()
> > alternative_instructions()
> >
> > So the first invocation of cpu_init() on the boot CPU will then use
> > static_cpu_has() which is not yet initialized proper.
>
> Ouch.
>
> Is there a way to catch such improper static_cpu_has() users?
> Silent misbehaviour is risky.
Yes, it is. I don't think we have something in place right now, but we
should add it definitely. PeterZ ????
> > So, no. That does not work and the proper fix is:
> >
> > -unsigned int __pgtable_l5_enabled __initdata;
> > +unsigned int __pgtable_l5_enabled __ro_after_init;
> >
> > and make cpu/common.c use the early variant. The extra 4 bytes storage are
> > not a problem and cpu_init() is not a fast path at all.
>
> Okay, I'll prepare the patch.
>
> BTW, if we go this path after all, shouldn't we revert these:
>
> 046c0dbec023 ("x86: Mark native_set_p4d() as __always_inline")
> 1ea66554d3b0 ("x86/mm: Mark p4d_offset() __always_inline")
In principle the always inline is fine, but the changelogs are quite
misleading and I really regret that I did not take the time to analyse that
proper when I applied the patches. At least we have catched it now.
So yes, please send the reverts along. Can you please add a proper root
cause analysis for the issues Arnd has observed to the changelogs so we
have it documented for later.
Thanks,
tglx
