percpu_ref: remove unnecessary ACCESS_ONCE() in percpu_ref_tryget_live()

__ref_is_percpu() needs the implied ACCESS_ONCE() in
lockless_dereference() on @ref->percpu_count_ptr because the value is
tested for !__PERCPU_REF_ATOMIC, which may be set asynchronously, and
then used as a pointer.  If the compiler generates a separate fetch
when using it as a pointer, __PERCPU_REF_ATOMIC may be set in between
contaminating the pointer value.

percpu_ref_tryget_live() also uses ACCESS_ONCE() to test
__PERCPU_REF_DEAD; however, there's no reason for this.  I just copied
ACCESS_ONCE() usage blindly from __ref_is_percpu().  All it does is
confusing people trying to understand what's going on.

This patch removes the unnecessary ACCESS_ONCE() usage from
percpu_ref_tryget_live() and adds a comment explaining why
__ref_is_percpu() needs it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Kent Overstreet <kmo@daterainc.com>

1 files changed, 17 insertions, 3 deletions

diff --git a/include/linux/percpu-refcount.h b/include/linux/percpu-refcount.h
index b4337646388b..6a7a670366ab 100644
--- a/include/linux/percpu-refcount.h
+++ b/include/linux/percpu-refcount.h
@@ -128,8 +128,22 @@ static inline void percpu_ref_kill(struct percpu_ref *ref)
 static inline bool __ref_is_percpu(struct percpu_ref *ref,
                                           unsigned long __percpu **percpu_countp)
 {
-        /* paired with smp_store_release() in percpu_ref_reinit() */
-        unsigned long percpu_ptr = lockless_dereference(ref->percpu_count_ptr);
+        unsigned long percpu_ptr;
+
+        /*
+         * The value of @ref->percpu_count_ptr is tested for
+         * !__PERCPU_REF_ATOMIC, which may be set asynchronously, and then
+         * used as a pointer.  If the compiler generates a separate fetch
+         * when using it as a pointer, __PERCPU_REF_ATOMIC may be set in
+         * between contaminating the pointer value, meaning that
+         * ACCESS_ONCE() is required when fetching it.
+         *
+         * Also, we need a data dependency barrier to be paired with
+         * smp_store_release() in __percpu_ref_switch_to_percpu().
+         *
+         * Use lockless deref which contains both.
+         */
+        percpu_ptr = lockless_dereference(ref->percpu_count_ptr);
 
         /*
          * Theoretically, the following could test just ATOMIC; however,
@@ -233,7 +247,7 @@ static inline bool percpu_ref_tryget_live(struct percpu_ref *ref)
         if (__ref_is_percpu(ref, &percpu_count)) {
                 this_cpu_inc(*percpu_count);
                 ret = true;
-        } else if (!(ACCESS_ONCE(ref->percpu_count_ptr) & __PERCPU_REF_DEAD)) {
+        } else if (!(ref->percpu_count_ptr & __PERCPU_REF_DEAD)) {
                 ret = atomic_long_inc_not_zero(&ref->count);
         }