Messages in this thread | | | Date | Wed, 9 Oct 2019 19:41:24 +0200 | From | Peter Zijlstra <> | Subject | Re: [PATCH v3 05/10] lib/refcount: Improve performance of generic REFCOUNT_FULL code |
| |
On Wed, Oct 09, 2019 at 05:44:33PM +0100, Will Deacon wrote:
> > > @@ -224,26 +208,19 @@ static inline void refcount_inc(refcount_t *r) > > > */ > > > static inline __must_check bool refcount_sub_and_test(int i, refcount_t *r) > > > { > > > + int old = atomic_fetch_sub_release(i, &r->refs); > > > > > > + if (old == i) { > > > smp_acquire__after_ctrl_dep(); > > > return true; > > > } > > > > > > + if (unlikely(old - i < 0)) { > > > + refcount_set(r, REFCOUNT_SATURATED); > > > + WARN_ONCE(1, "refcount_t: underflow; use-after-free.\n"); > > > + } > > > > I'm failing to see how this preserves REFCOUNT_SATURATED for > > non-underflow. AFAICT this should have: > > > > if (unlikely(old == REFCOUNT_SATURATED || old - i < 0)) > > Well spotted! I think we just want: > > if (unlikely(old < 0 || old - i < 0)) > > here, which is reassuringly similar to the logic in refcount_add() and > refcount_add_not_zero().
Oh indeed, I missed that saturated was negative. That should work.
> > > + return false; > > > } > > > > > > /** > > > @@ -276,9 +253,13 @@ static inline __must_check bool refcount_dec_and_test(refcount_t *r) > > > */ > > > static inline void refcount_dec(refcount_t *r) > > > { > > > + int old = atomic_fetch_sub_release(1, &r->refs); > > > > > > + if (unlikely(old <= 1)) { > > > > Idem. > > Hmm, I don't get what you mean with the one, since we're looking at the > old value. REFCOUNT_SATURATED is negative, so it will do the right thing.
Yep, missed that.
> > > + refcount_set(r, REFCOUNT_SATURATED); > > > + WARN_ONCE(1, "refcount_t: decrement hit 0; leaking memory.\n"); > > > + } > > > +} > > > > Also, things like refcount_dec_not_one() might need fixing to preserve > > REFCOUNT_SATURATED, because they're not expecting that value to actually > > change, but you do! > > refcount_dec_not_one() already checks for REFCOUNT_SATURATED and, in the > case of a racing thread setting the saturated value, the cmpxchg() will > fail if the saturated value is written after the check or the saturated > value will overwrite the value written by the cmpxchg(). Is there another > race that you're thinking of?
Hmm, yes. I was afraid that by not recognising SATURATED it'd go wrong, but now that I try I can't make it go wrong.
| |