Re: PREEMPT_RCU breaks anon_vma locking ?

[Hugh Dickins <hugh@veritas.com>]

On Fri, 23 Feb 2007, Paul E. McKenney wrote:
> 
> This look like a valid fix to me, at least as long as the lock is never
> dropped in the meantime (e.g., to do I/O).  If the lock -is- dropped in
> the meantime, then presumably whatever is done to keep the page from
> vanishing should allow an rcu_read_unlock() to be placed after each
> spin_unlock(&...->lock) and an rcu_read_lock() to be placed before each
> spin_lock(&...->lock).

Thankfully no complications of that kind, page_lock_anon_vma is static
to mm/rmap.c, and only used to hold the spin lock while examining page
tables of the vmas in the list, never a need to drop that lock at all.
(Until the day when someone reports such a long list that we start to
worry about the latency.)

Hugh