Messages in this thread | | | Subject | Re: Q. locking order of dcache_lru_lock | From | Peter Zijlstra <> | Date | Mon, 11 Apr 2011 10:30:51 +0200 |
| |
On Mon, 2011-04-11 at 14:09 +0900, J. R. Okajima wrote: > Peter Zijlstra: > > On Fri, 2011-04-08 at 22:20 +0900, J. R. Okajima wrote: > > >=20 > > > When spin_trylock(&dentry->d_lock) successfully acquired d_lock, does > > > the violation of locking order happen (or a deadlock, in worse case)?=20 > > > > No, since a trylock never actually blocks a deadlock cannot occur. > > Ah, exactly. I had to be sleeping when I wrote about deadlock. > How about the locking order? Do you think d_lock after dcache_lru_lock > is a problem?
Not really a problem, locking order is simply a tool/scheme to avoid deadlocks. Since there is no deadlock potential its fine to 'violate' locking order.
This is a common pattern with trylocks. In situations where you would need somewhat expensive lock operations to grab the locks in the right order, you can trylock to see if you can get them in the wrong order. If the trylock succeeds, yay! you got it cheap. If not, bummer, and you have to try the expensive way.
| |