sys_membarrier() scheduler barrier requirements: dropped

From: Paul E. McKenney
Date: Mon Sep 18 2017 - 13:08:04 EST


Hello!

Commit 3ed668659e95 ("membarrier: Document scheduler barrier
requirements") did not make it into the v4.14 merge window, and rebasing
to v4.14-rc1 results in conflicts. I have therefore dropped it.
If someone would be willing to forward-port it, I would be quite happy
to pull it back in for the v4.15 merge window. Or potentially as part
of the fix to sys_membarrier() for 4.14, for that matter.

Thanx, Paul

------------------------------------------------------------------------

commit 3ed668659e95ecfb6f6be0a3e7ff0fa6d27b2f5c
Author: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
Date: Fri Aug 18 21:39:16 2017 -0700

membarrier: Document scheduler barrier requirements

Document the membarrier requirement on having a full memory barrier in
__schedule() after coming from user-space, before storing to rq->curr.
It is provided by smp_mb__before_spinlock() in __schedule().

Document that membarrier requires a full barrier on transition from
kernel thread to userspace thread, which skips the call to switch_mm(). We
currently have an implicit barrier from atomic_dec_and_test() in mmdrop() that
ensures this.

The x86 switch_mm_irqs_off() full barrier is currently provided by many cpumask
update operations as well as load_cr3(). Document that load_cr3() is providing
this barrier.

[ Rebased on top of linux-rcu for-mingo branch.
Applies on top of "membarrier: Provide expedited private command". ]

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx>
CC: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
CC: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
CC: Boqun Feng <boqun.feng@xxxxxxxxx>
CC: Andrew Hunter <ahh@xxxxxxxxxx>
CC: Maged Michael <maged.michael@xxxxxxxxx>
CC: gromer@xxxxxxxxxx
CC: Avi Kivity <avi@xxxxxxxxxxxx>
CC: Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx>
CC: Paul Mackerras <paulus@xxxxxxxxx>
CC: Michael Ellerman <mpe@xxxxxxxxxxxxxx>
CC: Dave Watson <davejwatson@xxxxxx>
Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>

diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index 014d07a80053..cd815b63420a 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -133,6 +133,9 @@ void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
* and neither LOCK nor MFENCE orders them.
* Fortunately, load_cr3() is serializing and gives the
* ordering guarantee we need.
+ *
+ * This full barrier is also required by the membarrier
+ * system call.
*/
load_cr3(next->pgd);

diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index 2b24a6974847..fe29d06e2800 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -38,6 +38,10 @@ static inline void mmgrab(struct mm_struct *mm)
extern void __mmdrop(struct mm_struct *);
static inline void mmdrop(struct mm_struct *mm)
{
+ /*
+ * The implicit full barrier implied by atomic_dec_and_test is
+ * required by the membarrier system call.
+ */
if (unlikely(atomic_dec_and_test(&mm->mm_count)))
__mmdrop(mm);
}
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 3f29c6a89d80..b0f199f9ec62 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2654,6 +2654,12 @@ static struct rq *finish_task_switch(struct task_struct *prev)
finish_arch_post_lock_switch();

fire_sched_in_preempt_notifiers(current);
+ /*
+ * When transitioning from a kernel thread to a userspace
+ * thread, mmdrop()'s implicit full barrier is required by the
+ * membarrier system call, because the current active_mm can
+ * become the current mm without going through switch_mm().
+ */
if (mm)
mmdrop(mm);
if (unlikely(prev_state == TASK_DEAD)) {
@@ -3295,6 +3301,9 @@ static void __sched notrace __schedule(bool preempt)
* Make sure that signal_pending_state()->signal_pending() below
* can't be reordered with __set_current_state(TASK_INTERRUPTIBLE)
* done by the caller to avoid the race with signal_wake_up().
+ *
+ * The membarrier system call requires a full memory barrier
+ * after coming from user-space, before storing to rq->curr.
*/
smp_mb__before_spinlock();
rq_lock(rq, &rf);