Re: [PATCH] lockdep: lock_set_subclass - reset a held lock's subclass

[Jeremy Fitzhardinge <jeremy@goop.org>]

Hugh Dickins wrote:
> Please check the spin_lock_nested() in move_ptes() in mm/mremap.c.
>
> If you have down_write(&mm->mmap_sem) then you should be safe,
> but may need to do something to placate lockdep.  If you don't
> have down_write(&mm->mmap_sem), then I think you're in trouble?
>
> Not a big deal, the move_ptes() locking can be adjusted to suit
> your rule, it was just easier to do it the way it is at the time.

Ah, yes, I did look at that.  I think it isn't an issue, because my code 
is called from dup_mmap(), activate_mm() or exit_mmap().

dup_mmap() already holds mmap_sem.

activate_mm() in exec doesn't hold the sem, but I don't think it's 
possible for anyone to be racing against it.
activate_mm() in unshare doesn't seem to get used.

exit_mmap() gets called when there are no other users, so we'd better 
not be racing...

    J