Re: [PATCH 10/10] mm/hugetlb: Introduce hugetlb_walk()

[Peter Xu]

 On Tue, Nov 29, 2022 at 09:18:08PM -0800, Eric Biggers wrote:
> On Tue, Nov 29, 2022 at 02:35:26PM -0500, Peter Xu wrote:
> > +static inline pte_t *
> > +hugetlb_walk(struct vm_area_struct *vma, unsigned long addr, unsigned long sz)
> > +{
> > +#if defined(CONFIG_ARCH_WANT_HUGE_PMD_SHARE) && defined(CONFIG_LOCKDEP)
> > +	struct hugetlb_vma_lock *vma_lock = vma->vm_private_data;
> > +
> > +	/*
> > +	 * If pmd sharing possible, locking needed to safely walk the
> > +	 * hugetlb pgtables.  More information can be found at the comment
> > +	 * above huge_pte_offset() in the same file.
> > +	 *
> > +	 * NOTE: lockdep_is_held() is only defined with CONFIG_LOCKDEP.
> > +	 */
> > +	if (__vma_shareable_flags_pmd(vma))
> > +		WARN_ON_ONCE(!lockdep_is_held(&vma_lock->rw_sema) &&
> > +			     !lockdep_is_held(
> > +				 &vma->vm_file->f_mapping->i_mmap_rwsem));
> > +#endif
> 
> FYI, in next-20221130 there is a compile error here due to this commit:
> 
>     In file included from security/commoncap.c:19:
>     ./include/linux/hugetlb.h:1262:42: error: incomplete definition of type 'struct hugetlb_vma_lock'
>                     WARN_ON_ONCE(!lockdep_is_held(&vma_lock->rw_sema) &&
>                                                    ~~~~~~~~^
>     ./include/linux/lockdep.h:286:47: note: expanded from macro 'lockdep_is_held'
>     #define lockdep_is_held(lock)           lock_is_held(&(lock)->dep_map)
>                                                            ^~~~

This probably means the config has:

  CONFIG_HUGETLB_PAGE=n
  CONFIG_ARCH_WANT_HUGE_PMD_SHARE=y

And I'm surprised we didn't have a dependency that ARCH_WANT_HUGE_PMD_SHARE
should depend on HUGETLB_PAGE already.  Mike, what do you think?

I've also attached a quick fix for this patch to be squashed in.  Hope it
works.

Thanks,

-- 
Peter Xu