Re: [PATCH 1/3] mm: Make generic arch_is_kernel_initmem_freed() do what it says

[Christophe Leroy]

Le 27/09/2021 à 15:11, Michael Ellerman a écrit :
> Christophe Leroy <christophe.leroy@xxxxxxxxxx> writes:
> > Commit 7a5da02de8d6 ("locking/lockdep: check for freed initmem in
> > static_obj()") added arch_is_kernel_initmem_freed() which is supposed
> > to report whether an object is part of already freed init memory.
> >
> > For the time being, the generic version of arch_is_kernel_initmem_freed()
> > always reports 'false', allthough free_initmem() is generically called
> > on all architectures.
> >
> > Therefore, change the generic version of arch_is_kernel_initmem_freed()
> > to check whether free_initmem() has been called. If so, then check
> > if a given address falls into init memory.
> >
> > In order to use function init_section_contains(), the fonction is
> > moved at the end of asm-generic/section.h
> >
> > Cc: Gerald Schaefer <gerald.schaefer@xxxxxxxxxxxxx>
> > Signed-off-by: Christophe Leroy <christophe.leroy@xxxxxxxxxx>
> > ---
> >   include/asm-generic/sections.h | 31 +++++++++++++++++--------------
> >   1 file changed, 17 insertions(+), 14 deletions(-)
> >
> > diff --git a/include/asm-generic/sections.h b/include/asm-generic/sections.h
> > index d16302d3eb59..d1e5bb2c6b72 100644
> > --- a/include/asm-generic/sections.h
> > +++ b/include/asm-generic/sections.h
> > @@ -172,4 +158,21 @@ static inline bool is_kernel_rodata(unsigned long addr)
> >   	       addr < (unsigned long)__end_rodata;
> >   }
> >   
> > +/*
> > + * Check if an address is part of freed initmem. This is needed on architectures
> > + * with virt == phys kernel mapping, for code that wants to check if an address
> > + * is part of a static object within [_stext, _end]. After initmem is freed,
> > + * memory can be allocated from it, and such allocations would then have
> > + * addresses within the range [_stext, _end].
> > + */
> > +#ifndef arch_is_kernel_initmem_freed
> > +static inline int arch_is_kernel_initmem_freed(unsigned long addr)
> > +{
> > +	if (system_state < SYSTEM_RUNNING)
> > +		return 0;
> > +
> > +	return init_section_contains((void *)addr, 1);
> > +}
> > +#endif
>
> This will return an incorrect result for a short period during boot
> won't it?
>
> See init/main.c:
>
> static int __ref kernel_init(void *unused)
> {
> 	...
> 	free_initmem();			<- memory is freed here
> 	mark_readonly();
>
> 	/*
> 	 * Kernel mappings are now finalized - update the userspace page-table
> 	 * to finalize PTI.
> 	 */
> 	pti_finalize();
>
> 	system_state = SYSTEM_RUNNING;
>
>
> After free_initmem() we have address ranges that are now freed initmem,
> but arch_is_kernel_initmem_freed() continues to return 0 (false) for all
> addresses, until we update system_state.
>
> Possibly that doesn't matter for any of the current callers, but it
> seems pretty dicey to me.

Yes I saw it but as function core_kernel_text() uses that criteria for 
deciding whether a given init text address is valid or not, I thought it 
was just ok.

Should we add an intermediate state called for exemple 
SYSTEM_FREEING_INIT just before SYSTEM_RUNNING ?

Christophe