percpu-refcount: use RCU-sched insted of normal RCU
Subsystems (2)
Tejun Heo
maintainerPATCH 1/1 percpu-refcount: use RCU-sched insted of normal RCU
percpu-refcount was incorrectly using preempt_disable/enable() for RCU critical sections against call_rcu(). 6a24474da8 ("percpu-refcount:
consistently use plain (non-sched) RCU") fixed it by converting the preepmtion operations with rcu_read_[un]lock() citing that there isn't any advantage in using sched-RCU over using the usual one; however, rcu_read_[un]lock() for the preemptible RCU implementation - CONFIG_TREE_PREEMPT_RCU, chosen when CONFIG_PREEMPT - are slightly more expensive than preempt_disable/enable().
In a contrived microbench which repeats the followings,
- percpu_ref_get()
- copy 32 bytes of data into percpu buffer
- percpu_put_get()
- copy 32 bytes of data into percpu buffer
rcu_read_[un]lock() used in percpu_ref_get/put() makes it go slower by about 15% when compared to using sched-RCU.
As the RCU critical sections are extremely short, using sched-RCU shouldn't have any latency implications. Convert to RCU-sched.
--- include/linux/percpu-refcount.h | 12 ++++++------ lib/percpu-refcount.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)
include/linux/percpu-refcount.h +6 -6
@@ -105,7 +105,7 @@ static inline void percpu_ref_get(struct {unsigned __percpu *pcpu_count; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);
@@ -114,7 +114,7 @@ static inline void percpu_ref_get(structelse atomic_inc (&ref->count); -rcu_read_unlock (); +rcu_read_unlock_sched (); }/**
@@ -134,7 +134,7 @@ static inline bool percpu_ref_tryget(strunsigned __percpu *pcpu_count;int ret = false; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);
@@ -143,7 +143,7 @@ static inline bool percpu_ref_tryget(str ret = true; } -rcu_read_unlock (); +rcu_read_unlock_sched ();return ret; }
@@ -159,7 +159,7 @@ static inline void percpu_ref_put(struct {unsigned __percpu *pcpu_count; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);
@@ -168,7 +168,7 @@ static inline void percpu_ref_put(struct else if (unlikely(atomic_dec_and_test(&ref->count))) ref->release(ref); - rcu_read_unlock(); + rcu_read_unlock_sched(); }#endif
lib/percpu-refcount.c +1 -1
@@ -154,5 +154,5 @@ void percpu_ref_kill_and_confirm(struct (((unsigned long ) ref->pcpu_count)|PCPU_REF_DEAD ); ref->confirm_kill = confirm_kill; -call_rcu (&ref->rcu, percpu_ref_kill_rcu); +call_rcu_sched (&ref->rcu, percpu_ref_kill_rcu); }
Quoted 19 lines · Tejun Heo
percpu-refcount was incorrectly using preempt_disable/enable() for RCU critical sections against
call_rcu().6a24474da8("percpu-refcount:
consistently use plain (non-sched) RCU") fixed it by converting the preepmtion operations with rcu_read_[un]lock() citing that there isn't any advantage in using sched-RCU over using the usual one; however, rcu_read_[un]lock() for the preemptible RCU implementation -CONFIG_TREE_PREEMPT_RCU, chosen whenCONFIG_PREEMPT- are slightly more expensive than preempt_disable/enable().In a contrived microbench which repeats the followings,
-
percpu_ref_get()
- copy 32 bytes of data into percpu buffer
-percpu_put_get()
- copy 32 bytes of data into percpu bufferrcu_read_[un]lock() used in percpu_ref_get/put() makes it go slower by about 15% when compared to using sched-RCU.
Quoted 7 lines
As the RCU critical sections are extremely short, using sched-RCU shouldn't have any latency implications. Convert to RCU-sched.
Signed-off-by Tejun Heo <[email protected]>Cc Kent Overstreet <[email protected]>Cc Michal Hocko <[email protected]>Cc "Paul E. McKenney" <[email protected]>
Quoted 72 lines
Cc Rusty Russell <[email protected]>--- include/linux/percpu-refcount.h | 12 ++++++------ lib/percpu-refcount.c | 2 +- 2 files changed, 7 insertions(+), 7 deletions(-)include/linux/percpu-refcount.h +6
-6@@ -105,7 +105,7 @@ static inline void percpu_ref_get(struct {unsigned __percpu *pcpu_count; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);@@ -114,7 +114,7 @@ static inline void percpu_ref_get(structelse atomic_inc (&ref->count); -rcu_read_unlock (); +rcu_read_unlock_sched (); }/** @@ -134,7 +134,7 @@ static inline bool percpu_ref_tryget(strunsigned __percpu *pcpu_count;int ret = false; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);@@ -143,7 +143,7 @@ static inline bool percpu_ref_tryget(str ret = true; } -rcu_read_unlock (); +rcu_read_unlock_sched ();return ret; }@@ -159,7 +159,7 @@ static inline void percpu_ref_put(struct {unsigned __percpu *pcpu_count; -rcu_read_lock (); +rcu_read_lock_sched (); pcpu_count =ACCESS_ONCE (ref->pcpu_count);@@ -168,7 +168,7 @@ static inline void percpu_ref_put(struct else if (unlikely(atomic_dec_and_test(&ref->count))) ref->release(ref); - rcu_read_unlock(); + rcu_read_unlock_sched(); }#endif lib/percpu-refcount.c +1
-1@@ -154,5 +154,5 @@ void percpu_ref_kill_and_confirm(struct (((unsigned long ) ref->pcpu_count)|PCPU_REF_DEAD ); ref->confirm_kill = confirm_kill; -call_rcu (&ref->rcu, percpu_ref_kill_rcu); +call_rcu_sched (&ref->rcu, percpu_ref_kill_rcu); }
Quoted 19 lines · Tejun Heo
percpu-refcount was incorrectly using preempt_disable/enable() for RCU critical sections against
call_rcu().6a24474da8("percpu-refcount:
consistently use plain (non-sched) RCU") fixed it by converting the preepmtion operations with rcu_read_[un]lock() citing that there isn't any advantage in using sched-RCU over using the usual one; however, rcu_read_[un]lock() for the preemptible RCU implementation -CONFIG_TREE_PREEMPT_RCU, chosen whenCONFIG_PREEMPT- are slightly more expensive than preempt_disable/enable().In a contrived microbench which repeats the followings,
-
percpu_ref_get()
- copy 32 bytes of data into percpu buffer
-percpu_put_get()
- copy 32 bytes of data into percpu bufferrcu_read_[un]lock() used in percpu_ref_get/put() makes it go slower by about 15% when compared to using sched-RCU.
Quoted 8 lines
As the RCU critical sections are extremely short, using sched-RCU shouldn't have any latency implications. Convert to RCU-sched.
Signed-off-by Tejun Heo <[email protected]>Cc Kent Overstreet <[email protected]>Cc Michal Hocko <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Rusty Russell <[email protected]>
Quoted 19 lines · Tejun Heo
percpu-refcount was incorrectly using preempt_disable/enable() for RCU critical sections against
call_rcu().6a24474da8("percpu-refcount:
consistently use plain (non-sched) RCU") fixed it by converting the preepmtion operations with rcu_read_[un]lock() citing that there isn't any advantage in using sched-RCU over using the usual one; however, rcu_read_[un]lock() for the preemptible RCU implementation -CONFIG_TREE_PREEMPT_RCU, chosen whenCONFIG_PREEMPT- are slightly more expensive than preempt_disable/enable().In a contrived microbench which repeats the followings,
-
percpu_ref_get()
- copy 32 bytes of data into percpu buffer
-percpu_put_get()
- copy 32 bytes of data into percpu bufferrcu_read_[un]lock() used in percpu_ref_get/put() makes it go slower by about 15% when compared to using sched-RCU.
Quoted 8 lines
As the RCU critical sections are extremely short, using sched-RCU shouldn't have any latency implications. Convert to RCU-sched.
Signed-off-by Tejun Heo <[email protected]>Cc Kent Overstreet <[email protected]>Cc Michal Hocko <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Rusty Russell <[email protected]>
Applied to percpu/for-3.11 with acks added. Thanks!
Signature · 1 line
tejun