On Mon, May 18, 2020 at 01:01:33PM +0200, Vlastimil Babka wrote:
> On 4/24/20 3:38 AM, Michel Lespinasse wrote:
> > +static inline void mmap_assert_locked(struct mm_struct *mm)
> > +{
> > + VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, -1), mm);
> > + VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> > +}
> > +
> > +static inline void mmap_assert_write_locked(struct mm_struct *mm)
> > +{
> > + VM_BUG_ON_MM(!lockdep_is_held_type(&mm->mmap_sem, 0), mm);
> > + VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
> > +}
>
> I would remove VM_BUG_ON_MM() from the lockdep part. If kernel has lockdep
> enabled, it's already in heavy debugging mode enough so let's just use it and
> not depend on DEBUG_VM. Many sites you convert don't require DEBUG_VM for the
> lockdep checks.
>
> With that you can also use the standard lockdep_assert_held() and
> lockdep_assert_held_write() wrappers.
>
> If user has both lockdep and DEBUG_VM enabled, should we run both variants?
> Perhaps lockdep is enough as it's more comprehensive? Your initial v5 version
> was doing that.
Thanks, changed these to lockdep_assert_held() /
lockdep_assert_held_write() as suggested. This misses dumping out the
mm, but I think there is only limited value in that. I did keep the
rwsem_is_locked fallback as people had commented earlier about
getting assertions in the non-lockdep case. If both are enabled...
then we'll get somewhat redundant assertions, but I think that is fine
(better to keep the code simple than try to work around that).
--
Michel "Walken" Lespinasse
A program is never fully debugged until the last user dies.
