[PATCH 0/3] Membarrier updates

[PATCH 0/3] Membarrier updates

[Mathieu Desnoyers]

Please find the following membarrier updates series posted for inclusion
upstream.

Thanks,

Mathieu

Mathieu Desnoyers (3):
  sched: fix exit_mm vs membarrier (v4)
  sched: membarrier: cover kthread_use_mm (v4)
  sched: membarrier: document memory ordering scenarios

 include/linux/sched/mm.h  |   5 ++
 kernel/exit.c             |  16 ++++-
 kernel/kthread.c          |  21 ++++++
 kernel/sched/idle.c       |   1 +
 kernel/sched/membarrier.c | 147 ++++++++++++++++++++++++++++++++++++--
 5 files changed, 185 insertions(+), 5 deletions(-)

-- 
2.17.1

[PATCH 1/3] sched: fix exit_mm vs membarrier (v4)

[Mathieu Desnoyers]

exit_mm should issue memory barriers after user-space memory accesses,
before clearing current->mm, to order user-space memory accesses
performed prior to exit_mm before clearing tsk->mm, which has the
effect of skipping the membarrier private expedited IPIs.

exit_mm should also update the runqueue's membarrier_state so
membarrier global expedited IPIs are not sent when they are not
needed.

The membarrier system call can be issued concurrently with do_exit
if we have thread groups created with CLONE_VM but not CLONE_THREAD.

Here is the scenario I have in mind:

Two thread groups are created, A and B. Thread group B is created by
issuing clone from group A with flag CLONE_VM set, but not CLONE_THREAD.
Let's assume we have a single thread within each thread group (Thread A
and Thread B).

The AFAIU we can have:

Userspace variables:

int x = 0, y = 0;

CPU 0                   CPU 1
Thread A                Thread B
(in thread group A)     (in thread group B)

x = 1
barrier()
y = 1
exit()
exit_mm()
current->mm = NULL;
                        r1 = load y
                        membarrier()
                          skips CPU 0 (no IPI) because its current mm is NULL
                        r2 = load x
                        BUG_ON(r1 == 1 && r2 == 0)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: linux-mm@kvack.org
---
Changes since v1:
- Use smp_mb__after_spinlock rather than smp_mb.
- Document race scenario in commit message.

Changes since v2:
- Introduce membarrier_update_current_mm,
- Use membarrier_update_current_mm to update rq's membarrier_state from
  exit_mm.

Changes since v3:
- Disable interrupts around call to membarrier_update_current_mm, which
  is required to access the runqueue's fields.
---
 include/linux/sched/mm.h  |  5 +++++
 kernel/exit.c             | 16 +++++++++++++++-
 kernel/sched/membarrier.c | 12 ++++++++++++
 3 files changed, 32 insertions(+), 1 deletion(-)

diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index f889e332912f..5dd7f56baaba 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -370,6 +370,8 @@ static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 
 extern void membarrier_exec_mmap(struct mm_struct *mm);
 
+extern void membarrier_update_current_mm(struct mm_struct *next_mm);
+
 #else
 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
@@ -384,6 +386,9 @@ static inline void membarrier_exec_mmap(struct mm_struct *mm)
 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 {
 }
+static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
+{
+}
 #endif
 
 #endif /* _LINUX_SCHED_MM_H */
diff --git a/kernel/exit.c b/kernel/exit.c
index 733e80f334e7..18ca74c07085 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -475,10 +475,24 @@ static void exit_mm(void)
 	BUG_ON(mm != current->active_mm);
 	/* more a memory barrier than a real lock */
 	task_lock(current);
+	/*
+	 * When a thread stops operating on an address space, the loop
+	 * in membarrier_private_expedited() may not observe that
+	 * tsk->mm, and the loop in membarrier_global_expedited() may
+	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
+	 * rq->membarrier_state, so those would not issue an IPI.
+	 * Membarrier requires a memory barrier after accessing
+	 * user-space memory, before clearing tsk->mm or the
+	 * rq->membarrier_state.
+	 */
+	smp_mb__after_spinlock();
+	local_irq_disable();
 	current->mm = NULL;
-	mmap_read_unlock(mm);
+	membarrier_update_current_mm(NULL);
 	enter_lazy_tlb(mm, current);
+	local_irq_enable();
 	task_unlock(current);
+	mmap_read_unlock(mm);
 	mm_update_next_owner(mm);
 	mmput(mm);
 	if (test_thread_flag(TIF_MEMDIE))
diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index 168479a7d61b..8bc8b8a888b7 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -63,6 +63,18 @@ void membarrier_exec_mmap(struct mm_struct *mm)
 	this_cpu_write(runqueues.membarrier_state, 0);
 }
 
+void membarrier_update_current_mm(struct mm_struct *next_mm)
+{
+	struct rq *rq = this_rq();
+	int membarrier_state = 0;
+
+	if (next_mm)
+		membarrier_state = atomic_read(&next_mm->membarrier_state);
+	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
+		return;
+	WRITE_ONCE(rq->membarrier_state, membarrier_state);
+}
+
 static int membarrier_global_expedited(void)
 {
 	int cpu;
-- 
2.17.1

[PATCH 2/3] sched: membarrier: cover kthread_use_mm (v4)

[Mathieu Desnoyers]

Add comments and memory barrier to kthread_use_mm and kthread_unuse_mm
to allow the effect of membarrier(2) to apply to kthreads accessing
user-space memory as well.

Given that no prior kthread use this guarantee and that it only affects
kthreads, adding this guarantee does not affect user-space ABI.

Refine the check in membarrier_global_expedited to exclude runqueues
running the idle thread rather than all kthreads from the IPI cpumask.

Now that membarrier_global_expedited can IPI kthreads, the scheduler
also needs to update the runqueue's membarrier_state when entering lazy
TLB state.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
---
Changes since v1:
- Add WARN_ON_ONCE(current->mm) in play_idle_precise (PeterZ),
- Use smp_mb__after_spinlock rather than smp_mb after task_lock.

Changes since v2:
- Update the rq's membarrier state on kthread use/unuse mm,
- The scheduler must use membarrier_switch_mm for lazy TLB case as well,
  now that global expedited membarrier IPIs kthreads.

Changes since v3:
- Revert back to not using membarrier_switch_mm for lazy TLB case. This
  is made OK by ensuring that the global expedited membarrier only skips
  kthreads which have a NULL mm, so it only skips kthreads which are in
  lazy TLB mode.
---
 kernel/kthread.c          | 21 +++++++++++++++++++++
 kernel/sched/idle.c       |  1 +
 kernel/sched/membarrier.c |  7 +++----
 3 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/kernel/kthread.c b/kernel/kthread.c
index 3edaa380dc7b..a396734d31f3 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -1248,6 +1248,7 @@ void kthread_use_mm(struct mm_struct *mm)
 		tsk->active_mm = mm;
 	}
 	tsk->mm = mm;
+	membarrier_update_current_mm(mm);
 	switch_mm_irqs_off(active_mm, mm, tsk);
 	local_irq_enable();
 	task_unlock(tsk);
@@ -1255,8 +1256,19 @@ void kthread_use_mm(struct mm_struct *mm)
 	finish_arch_post_lock_switch();
 #endif
 
+	/*
+	 * When a kthread starts operating on an address space, the loop
+	 * in membarrier_{private,global}_expedited() may not observe
+	 * that tsk->mm, and not issue an IPI. Membarrier requires a
+	 * memory barrier after storing to tsk->mm, before accessing
+	 * user-space memory. A full memory barrier for membarrier
+	 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
+	 * mmdrop(), or explicitly with smp_mb().
+	 */
 	if (active_mm != mm)
 		mmdrop(active_mm);
+	else
+		smp_mb();
 
 	to_kthread(tsk)->oldfs = force_uaccess_begin();
 }
@@ -1276,9 +1288,18 @@ void kthread_unuse_mm(struct mm_struct *mm)
 	force_uaccess_end(to_kthread(tsk)->oldfs);
 
 	task_lock(tsk);
+	/*
+	 * When a kthread stops operating on an address space, the loop
+	 * in membarrier_{private,global}_expedited() may not observe
+	 * that tsk->mm, and not issue an IPI. Membarrier requires a
+	 * memory barrier after accessing user-space memory, before
+	 * clearing tsk->mm.
+	 */
+	smp_mb__after_spinlock();
 	sync_mm_rss(mm);
 	local_irq_disable();
 	tsk->mm = NULL;
+	membarrier_update_current_mm(NULL);
 	/* active_mm is still 'mm' */
 	enter_lazy_tlb(mm, tsk);
 	local_irq_enable();
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index f324dc36fc43..f0d81a5ea471 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -338,6 +338,7 @@ void play_idle_precise(u64 duration_ns, u64 latency_ns)
 	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
 	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
 	WARN_ON_ONCE(!duration_ns);
+	WARN_ON_ONCE(current->mm);
 
 	rcu_sleep_check();
 	preempt_disable();
diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index 8bc8b8a888b7..8b93b6844901 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -113,12 +113,11 @@ static int membarrier_global_expedited(void)
 			continue;
 
 		/*
-		 * Skip the CPU if it runs a kernel thread. The scheduler
-		 * leaves the prior task mm in place as an optimization when
-		 * scheduling a kthread.
+		 * Skip the CPU if it runs a kernel thread which is not using
+		 * a task mm.
 		 */
 		p = rcu_dereference(cpu_rq(cpu)->curr);
-		if (p->flags & PF_KTHREAD)
+		if (!p->mm)
 			continue;
 
 		__cpumask_set_cpu(cpu, tmpmask);
-- 
2.17.1

[PATCH 3/3] sched: membarrier: document memory ordering scenarios

[Mathieu Desnoyers]

Document membarrier ordering scenarios in membarrier.c. Thanks to Alan
Stern for refreshing my memory. Now that I have those in mind, it seems
appropriate to serialize them to comments for posterity.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Alan Stern <stern@rowland.harvard.edu>
---
 kernel/sched/membarrier.c | 128 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 128 insertions(+)

diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index 8b93b6844901..943bdf5e9108 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -6,6 +6,134 @@
  */
 #include "sched.h"
 
+/*
+ * For documentation purposes, here are some membarrier ordering
+ * scenarios to keep in mind:
+ *
+ * A) Userspace thread execution after IPI vs membarrier's memory
+ *    barrier before sending the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the start of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU0 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU1 after the IPI-induced memory barrier:
+ *
+ *         CPU0                              CPU1
+ *
+ *         x = 1
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r2 = y
+ *                                           y = 1
+ *                                           barrier()
+ *                                           r1 = x
+ *
+ *                     BUG_ON(r1 == 0 && r2 == 0)
+ *
+ * The write to y and load from x by CPU1 are unordered by the hardware,
+ * so it's possible to have "r1 = x" reordered before "y = 1" at any
+ * point after (b).  If the memory barrier at (a) is omitted, then "x = 1"
+ * can be reordered after (a) (although not after (c)), so we get r1 == 0
+ * and r2 == 0.  This violates the guarantee that membarrier() is
+ * supposed by provide.
+ *
+ * The timing of the memory barrier at (a) has to ensure that it executes
+ * before the IPI-induced memory barrier on CPU1.
+ *
+ * B) Userspace thread execution before IPI vs membarrier's memory
+ *    barrier after completing the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the end of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU1 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU0 after the membarrier():
+ *
+ *         CPU0                              CPU1
+ *
+ *                                           x = 1
+ *                                           barrier()
+ *                                           y = 1
+ *         r2 = y
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r1 = x
+ *         BUG_ON(r1 == 0 && r2 == 1)
+ *
+ * The writes to x and y are unordered by the hardware, so it's possible to
+ * have "r2 = 1" even though the write to x doesn't execute until (b).  If
+ * the memory barrier at (c) is omitted then "r1 = x" can be reordered
+ * before (b) (although not before (a)), so we get "r1 = 0".  This violates
+ * the guarantee that membarrier() is supposed to provide.
+ *
+ * The timing of the memory barrier at (c) has to ensure that it executes
+ * after the IPI-induced memory barrier on CPU1.
+ *
+ * C) Scheduling userspace thread -> kthread -> userspace thread vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           d: switch to kthread (includes mb)
+ *           b: read rq->curr->mm == NULL
+ *                                           e: switch to user (includes mb)
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (e). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ *
+ * D) exit_mm vs membarrier
+ *
+ * Two thread groups are created, A and B.  Thread group B is created by
+ * issuing clone from group A with flag CLONE_VM set, but not CLONE_THREAD.
+ * Let's assume we have a single thread within each thread group (Thread A
+ * and Thread B).  Thread A runs on CPU0, Thread B runs on CPU1.
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *             a: smp_mb()
+ *                                           exit_mm():
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *             b: read rq->curr->mm == NULL
+ *             c: smp_mb()
+ *
+ * Using scenario (B), we can show that (c) needs to be paired with (d).
+ *
+ * E) kthread_{use,unuse}_mm vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           kthread_unuse_mm()
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *           b: read rq->curr->mm == NULL
+ *                                           kthread_use_mm()
+ *                                             f: current->mm = mm
+ *                                             g: smp_mb()
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (g). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ */
+
 /*
  * Bitmask made from a "or" of all commands within enum membarrier_cmd,
  * except MEMBARRIER_CMD_QUERY.
-- 
2.17.1

Re: [PATCH 1/3] sched: fix exit_mm vs membarrier (v4)

[Peter Zijlstra]

On Tue, Oct 20, 2020 at 09:47:13AM -0400, Mathieu Desnoyers wrote:
> +void membarrier_update_current_mm(struct mm_struct *next_mm)
> +{
> +	struct rq *rq = this_rq();
> +	int membarrier_state = 0;
> +
> +	if (next_mm)
> +		membarrier_state = atomic_read(&next_mm->membarrier_state);
> +	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
> +		return;
> +	WRITE_ONCE(rq->membarrier_state, membarrier_state);
> +}

This is suspisioucly similar to membarrier_switch_mm().

Would something like so make sense?

---
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -206,14 +206,7 @@ void membarrier_exec_mmap(struct mm_stru
 
 void membarrier_update_current_mm(struct mm_struct *next_mm)
 {
-	struct rq *rq = this_rq();
-	int membarrier_state = 0;
-
-	if (next_mm)
-		membarrier_state = atomic_read(&next_mm->membarrier_state);
-	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
-		return;
-	WRITE_ONCE(rq->membarrier_state, membarrier_state);
+	membarrier_switch_mm(this_rq(), NULL, next_mm);
 }
 
 static int membarrier_global_expedited(void)
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index d2621155393c..3d589c2ffd28 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2645,12 +2645,14 @@ static inline void membarrier_switch_mm(struct rq *rq,
 					struct mm_struct *prev_mm,
 					struct mm_struct *next_mm)
 {
-	int membarrier_state;
+	int membarrier_state = 0;
 
 	if (prev_mm == next_mm)
 		return;
 
-	membarrier_state = atomic_read(&next_mm->membarrier_state);
+	if (next_mm)
+		membarrier_state = atomic_read(&next_mm->membarrier_state);
+
 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
 		return;
 

Re: [PATCH 1/3] sched: fix exit_mm vs membarrier (v4)

[Mathieu Desnoyers]

----- On Oct 20, 2020, at 10:36 AM, Peter Zijlstra peterz@infradead.org wrote:

> On Tue, Oct 20, 2020 at 09:47:13AM -0400, Mathieu Desnoyers wrote:
>> +void membarrier_update_current_mm(struct mm_struct *next_mm)
>> +{
>> +	struct rq *rq = this_rq();
>> +	int membarrier_state = 0;
>> +
>> +	if (next_mm)
>> +		membarrier_state = atomic_read(&next_mm->membarrier_state);
>> +	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
>> +		return;
>> +	WRITE_ONCE(rq->membarrier_state, membarrier_state);
>> +}
> 
> This is suspisioucly similar to membarrier_switch_mm().
> 
> Would something like so make sense?

Very much yes. Do you want me to re-send the series, or you
want to fold this in as you merge it ?

Thanks,

Mathieu

> 
> ---
> --- a/kernel/sched/membarrier.c
> +++ b/kernel/sched/membarrier.c
> @@ -206,14 +206,7 @@ void membarrier_exec_mmap(struct mm_stru
> 
> void membarrier_update_current_mm(struct mm_struct *next_mm)
> {
> -	struct rq *rq = this_rq();
> -	int membarrier_state = 0;
> -
> -	if (next_mm)
> -		membarrier_state = atomic_read(&next_mm->membarrier_state);
> -	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
> -		return;
> -	WRITE_ONCE(rq->membarrier_state, membarrier_state);
> +	membarrier_switch_mm(this_rq(), NULL, next_mm);
> }
> 
> static int membarrier_global_expedited(void)
> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index d2621155393c..3d589c2ffd28 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -2645,12 +2645,14 @@ static inline void membarrier_switch_mm(struct rq *rq,
> 					struct mm_struct *prev_mm,
> 					struct mm_struct *next_mm)
> {
> -	int membarrier_state;
> +	int membarrier_state = 0;
> 
> 	if (prev_mm == next_mm)
> 		return;
> 
> -	membarrier_state = atomic_read(&next_mm->membarrier_state);
> +	if (next_mm)
> +		membarrier_state = atomic_read(&next_mm->membarrier_state);
> +
> 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
>  		return;

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com

Re: [PATCH 1/3] sched: fix exit_mm vs membarrier (v4)

[Boqun Feng]

Hi,

On Tue, Oct 20, 2020 at 10:59:58AM -0400, Mathieu Desnoyers wrote:
> ----- On Oct 20, 2020, at 10:36 AM, Peter Zijlstra peterz@infradead.org wrote:
> 
> > On Tue, Oct 20, 2020 at 09:47:13AM -0400, Mathieu Desnoyers wrote:
> >> +void membarrier_update_current_mm(struct mm_struct *next_mm)
> >> +{
> >> +	struct rq *rq = this_rq();
> >> +	int membarrier_state = 0;
> >> +
> >> +	if (next_mm)
> >> +		membarrier_state = atomic_read(&next_mm->membarrier_state);
> >> +	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
> >> +		return;
> >> +	WRITE_ONCE(rq->membarrier_state, membarrier_state);
> >> +}
> > 
> > This is suspisioucly similar to membarrier_switch_mm().
> > 
> > Would something like so make sense?
> 
> Very much yes. Do you want me to re-send the series, or you
> want to fold this in as you merge it ?
> 
> Thanks,
> 
> Mathieu
> 
> > 
> > ---
> > --- a/kernel/sched/membarrier.c
> > +++ b/kernel/sched/membarrier.c
> > @@ -206,14 +206,7 @@ void membarrier_exec_mmap(struct mm_stru
> > 
> > void membarrier_update_current_mm(struct mm_struct *next_mm)
> > {
> > -	struct rq *rq = this_rq();
> > -	int membarrier_state = 0;
> > -
> > -	if (next_mm)
> > -		membarrier_state = atomic_read(&next_mm->membarrier_state);
> > -	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
> > -		return;
> > -	WRITE_ONCE(rq->membarrier_state, membarrier_state);
> > +	membarrier_switch_mm(this_rq(), NULL, next_mm);
> > }
> > 
> > static int membarrier_global_expedited(void)
> > diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> > index d2621155393c..3d589c2ffd28 100644
> > --- a/kernel/sched/sched.h
> > +++ b/kernel/sched/sched.h
> > @@ -2645,12 +2645,14 @@ static inline void membarrier_switch_mm(struct rq *rq,
> > 					struct mm_struct *prev_mm,
> > 					struct mm_struct *next_mm)
> > {
> > -	int membarrier_state;
> > +	int membarrier_state = 0;
> > 
> > 	if (prev_mm == next_mm)

Unless I'm missing something subtle, in exit_mm(),
membarrier_update_current_mm() is called with @next_mm == NULL, and
inside membarrier_update_current_mm(), membarrier_switch_mm() is called
wiht @prev_mm == NULL. As a result, the branch above is taken, so
membarrier_update_current_mm() becomes a nop. I think we should use the
previous value of current->mm as the @prev_mm, something like below
maybe?

void update_current_mm(struct mm_struct *next_mm)
{
	struct mm_struct *prev_mm;
	unsigned long flags;

	local_irq_save(flags);
	prev_mm = current->mm;
	current->mm = next_mm;
	membarrier_switch_mm(this_rq(), prev_mm, next_mm);
	local_irq_restore(flags);
}

, and replace all settings for "current->mm" in kernel with
update_current_mm().

Thoughts?

Regards,
Boqun

> > 		return;
> > 
> > -	membarrier_state = atomic_read(&next_mm->membarrier_state);
> > +	if (next_mm)
> > +		membarrier_state = atomic_read(&next_mm->membarrier_state);
> > +
> > 	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
> >  		return;
> 
> -- 
> Mathieu Desnoyers
> EfficiOS Inc.
> http://www.efficios.com

[tip: sched/core] sched: membarrier: cover kthread_use_mm (v4)

[tip-bot2 for Mathieu Desnoyers]

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     618758ed3a4f7d790414d020b362111748ebbf9f
Gitweb:        https://git.kernel.org/tip/618758ed3a4f7d790414d020b362111748ebbf9f
Author:        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
AuthorDate:    Tue, 20 Oct 2020 09:47:14 -04:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Thu, 29 Oct 2020 11:00:31 +01:00

sched: membarrier: cover kthread_use_mm (v4)

Add comments and memory barrier to kthread_use_mm and kthread_unuse_mm
to allow the effect of membarrier(2) to apply to kthreads accessing
user-space memory as well.

Given that no prior kthread use this guarantee and that it only affects
kthreads, adding this guarantee does not affect user-space ABI.

Refine the check in membarrier_global_expedited to exclude runqueues
running the idle thread rather than all kthreads from the IPI cpumask.

Now that membarrier_global_expedited can IPI kthreads, the scheduler
also needs to update the runqueue's membarrier_state when entering lazy
TLB state.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201020134715.13909-3-mathieu.desnoyers@efficios.com
---
 kernel/kthread.c          | 21 +++++++++++++++++++++
 kernel/sched/idle.c       |  1 +
 kernel/sched/membarrier.c |  7 +++----
 3 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/kernel/kthread.c b/kernel/kthread.c
index e29773c..481428f 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -1248,6 +1248,7 @@ void kthread_use_mm(struct mm_struct *mm)
 		tsk->active_mm = mm;
 	}
 	tsk->mm = mm;
+	membarrier_update_current_mm(mm);
 	switch_mm_irqs_off(active_mm, mm, tsk);
 	local_irq_enable();
 	task_unlock(tsk);
@@ -1255,8 +1256,19 @@ void kthread_use_mm(struct mm_struct *mm)
 	finish_arch_post_lock_switch();
 #endif
 
+	/*
+	 * When a kthread starts operating on an address space, the loop
+	 * in membarrier_{private,global}_expedited() may not observe
+	 * that tsk->mm, and not issue an IPI. Membarrier requires a
+	 * memory barrier after storing to tsk->mm, before accessing
+	 * user-space memory. A full memory barrier for membarrier
+	 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
+	 * mmdrop(), or explicitly with smp_mb().
+	 */
 	if (active_mm != mm)
 		mmdrop(active_mm);
+	else
+		smp_mb();
 
 	to_kthread(tsk)->oldfs = force_uaccess_begin();
 }
@@ -1276,9 +1288,18 @@ void kthread_unuse_mm(struct mm_struct *mm)
 	force_uaccess_end(to_kthread(tsk)->oldfs);
 
 	task_lock(tsk);
+	/*
+	 * When a kthread stops operating on an address space, the loop
+	 * in membarrier_{private,global}_expedited() may not observe
+	 * that tsk->mm, and not issue an IPI. Membarrier requires a
+	 * memory barrier after accessing user-space memory, before
+	 * clearing tsk->mm.
+	 */
+	smp_mb__after_spinlock();
 	sync_mm_rss(mm);
 	local_irq_disable();
 	tsk->mm = NULL;
+	membarrier_update_current_mm(NULL);
 	/* active_mm is still 'mm' */
 	enter_lazy_tlb(mm, tsk);
 	local_irq_enable();
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index 24d0ee2..846743e 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -338,6 +338,7 @@ void play_idle_precise(u64 duration_ns, u64 latency_ns)
 	WARN_ON_ONCE(!(current->flags & PF_KTHREAD));
 	WARN_ON_ONCE(!(current->flags & PF_NO_SETAFFINITY));
 	WARN_ON_ONCE(!duration_ns);
+	WARN_ON_ONCE(current->mm);
 
 	rcu_sleep_check();
 	preempt_disable();
diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index aac3292..f223f35 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -126,12 +126,11 @@ static int membarrier_global_expedited(void)
 			continue;
 
 		/*
-		 * Skip the CPU if it runs a kernel thread. The scheduler
-		 * leaves the prior task mm in place as an optimization when
-		 * scheduling a kthread.
+		 * Skip the CPU if it runs a kernel thread which is not using
+		 * a task mm.
 		 */
 		p = rcu_dereference(cpu_rq(cpu)->curr);
-		if (p->flags & PF_KTHREAD)
+		if (!p->mm)
 			continue;
 
 		__cpumask_set_cpu(cpu, tmpmask);

[tip: sched/core] sched: membarrier: document memory ordering scenarios

[tip-bot2 for Mathieu Desnoyers]

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     25595eb6aaa9fbb31330f1e0b400642694bc6574
Gitweb:        https://git.kernel.org/tip/25595eb6aaa9fbb31330f1e0b400642694bc6574
Author:        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
AuthorDate:    Tue, 20 Oct 2020 09:47:15 -04:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Thu, 29 Oct 2020 11:00:31 +01:00

sched: membarrier: document memory ordering scenarios

Document membarrier ordering scenarios in membarrier.c. Thanks to Alan
Stern for refreshing my memory. Now that I have those in mind, it seems
appropriate to serialize them to comments for posterity.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201020134715.13909-4-mathieu.desnoyers@efficios.com
---
 kernel/sched/membarrier.c | 128 +++++++++++++++++++++++++++++++++++++-
 1 file changed, 128 insertions(+)

diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index f223f35..5a40b38 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -7,6 +7,134 @@
 #include "sched.h"
 
 /*
+ * For documentation purposes, here are some membarrier ordering
+ * scenarios to keep in mind:
+ *
+ * A) Userspace thread execution after IPI vs membarrier's memory
+ *    barrier before sending the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the start of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU0 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU1 after the IPI-induced memory barrier:
+ *
+ *         CPU0                              CPU1
+ *
+ *         x = 1
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r2 = y
+ *                                           y = 1
+ *                                           barrier()
+ *                                           r1 = x
+ *
+ *                     BUG_ON(r1 == 0 && r2 == 0)
+ *
+ * The write to y and load from x by CPU1 are unordered by the hardware,
+ * so it's possible to have "r1 = x" reordered before "y = 1" at any
+ * point after (b).  If the memory barrier at (a) is omitted, then "x = 1"
+ * can be reordered after (a) (although not after (c)), so we get r1 == 0
+ * and r2 == 0.  This violates the guarantee that membarrier() is
+ * supposed by provide.
+ *
+ * The timing of the memory barrier at (a) has to ensure that it executes
+ * before the IPI-induced memory barrier on CPU1.
+ *
+ * B) Userspace thread execution before IPI vs membarrier's memory
+ *    barrier after completing the IPI
+ *
+ * Userspace variables:
+ *
+ * int x = 0, y = 0;
+ *
+ * The memory barrier at the end of membarrier() on CPU0 is necessary in
+ * order to enforce the guarantee that any writes occurring on CPU1 before
+ * the membarrier() is executed will be visible to any code executing on
+ * CPU0 after the membarrier():
+ *
+ *         CPU0                              CPU1
+ *
+ *                                           x = 1
+ *                                           barrier()
+ *                                           y = 1
+ *         r2 = y
+ *         membarrier():
+ *           a: smp_mb()
+ *           b: send IPI                       IPI-induced mb
+ *           c: smp_mb()
+ *         r1 = x
+ *         BUG_ON(r1 == 0 && r2 == 1)
+ *
+ * The writes to x and y are unordered by the hardware, so it's possible to
+ * have "r2 = 1" even though the write to x doesn't execute until (b).  If
+ * the memory barrier at (c) is omitted then "r1 = x" can be reordered
+ * before (b) (although not before (a)), so we get "r1 = 0".  This violates
+ * the guarantee that membarrier() is supposed to provide.
+ *
+ * The timing of the memory barrier at (c) has to ensure that it executes
+ * after the IPI-induced memory barrier on CPU1.
+ *
+ * C) Scheduling userspace thread -> kthread -> userspace thread vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           d: switch to kthread (includes mb)
+ *           b: read rq->curr->mm == NULL
+ *                                           e: switch to user (includes mb)
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (e). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ *
+ * D) exit_mm vs membarrier
+ *
+ * Two thread groups are created, A and B.  Thread group B is created by
+ * issuing clone from group A with flag CLONE_VM set, but not CLONE_THREAD.
+ * Let's assume we have a single thread within each thread group (Thread A
+ * and Thread B).  Thread A runs on CPU0, Thread B runs on CPU1.
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *             a: smp_mb()
+ *                                           exit_mm():
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *             b: read rq->curr->mm == NULL
+ *             c: smp_mb()
+ *
+ * Using scenario (B), we can show that (c) needs to be paired with (d).
+ *
+ * E) kthread_{use,unuse}_mm vs membarrier
+ *
+ *           CPU0                            CPU1
+ *
+ *           membarrier():
+ *           a: smp_mb()
+ *                                           kthread_unuse_mm()
+ *                                             d: smp_mb()
+ *                                             e: current->mm = NULL
+ *           b: read rq->curr->mm == NULL
+ *                                           kthread_use_mm()
+ *                                             f: current->mm = mm
+ *                                             g: smp_mb()
+ *           c: smp_mb()
+ *
+ * Using the scenario from (A), we can show that (a) needs to be paired
+ * with (g). Using the scenario from (B), we can show that (c) needs to
+ * be paired with (d).
+ */
+
+/*
  * Bitmask made from a "or" of all commands within enum membarrier_cmd,
  * except MEMBARRIER_CMD_QUERY.
  */

[tip: sched/core] sched: fix exit_mm vs membarrier (v4)

[tip-bot2 for Mathieu Desnoyers]

The following commit has been merged into the sched/core branch of tip:

Commit-ID:     5bc78502322a5e4eef3f1b2a2813751dc6434143
Gitweb:        https://git.kernel.org/tip/5bc78502322a5e4eef3f1b2a2813751dc6434143
Author:        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
AuthorDate:    Tue, 20 Oct 2020 09:47:13 -04:00
Committer:     Peter Zijlstra <peterz@infradead.org>
CommitterDate: Thu, 29 Oct 2020 11:00:30 +01:00

sched: fix exit_mm vs membarrier (v4)

exit_mm should issue memory barriers after user-space memory accesses,
before clearing current->mm, to order user-space memory accesses
performed prior to exit_mm before clearing tsk->mm, which has the
effect of skipping the membarrier private expedited IPIs.

exit_mm should also update the runqueue's membarrier_state so
membarrier global expedited IPIs are not sent when they are not
needed.

The membarrier system call can be issued concurrently with do_exit
if we have thread groups created with CLONE_VM but not CLONE_THREAD.

Here is the scenario I have in mind:

Two thread groups are created, A and B. Thread group B is created by
issuing clone from group A with flag CLONE_VM set, but not CLONE_THREAD.
Let's assume we have a single thread within each thread group (Thread A
and Thread B).

The AFAIU we can have:

Userspace variables:

int x = 0, y = 0;

CPU 0                   CPU 1
Thread A                Thread B
(in thread group A)     (in thread group B)

x = 1
barrier()
y = 1
exit()
exit_mm()
current->mm = NULL;
                        r1 = load y
                        membarrier()
                          skips CPU 0 (no IPI) because its current mm is NULL
                        r2 = load x
                        BUG_ON(r1 == 1 && r2 == 0)

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20201020134715.13909-2-mathieu.desnoyers@efficios.com
---
 include/linux/sched/mm.h  |  5 +++++
 kernel/exit.c             | 16 +++++++++++++++-
 kernel/sched/membarrier.c | 12 ++++++++++++
 3 files changed, 32 insertions(+), 1 deletion(-)

diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index d5ece7a..a91fb3a 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -347,6 +347,8 @@ static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 
 extern void membarrier_exec_mmap(struct mm_struct *mm);
 
+extern void membarrier_update_current_mm(struct mm_struct *next_mm);
+
 #else
 #ifdef CONFIG_ARCH_HAS_MEMBARRIER_CALLBACKS
 static inline void membarrier_arch_switch_mm(struct mm_struct *prev,
@@ -361,6 +363,9 @@ static inline void membarrier_exec_mmap(struct mm_struct *mm)
 static inline void membarrier_mm_sync_core_before_usermode(struct mm_struct *mm)
 {
 }
+static inline void membarrier_update_current_mm(struct mm_struct *next_mm)
+{
+}
 #endif
 
 #endif /* _LINUX_SCHED_MM_H */
diff --git a/kernel/exit.c b/kernel/exit.c
index 87a2d51..a3dd6b3 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -475,10 +475,24 @@ static void exit_mm(void)
 	BUG_ON(mm != current->active_mm);
 	/* more a memory barrier than a real lock */
 	task_lock(current);
+	/*
+	 * When a thread stops operating on an address space, the loop
+	 * in membarrier_private_expedited() may not observe that
+	 * tsk->mm, and the loop in membarrier_global_expedited() may
+	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
+	 * rq->membarrier_state, so those would not issue an IPI.
+	 * Membarrier requires a memory barrier after accessing
+	 * user-space memory, before clearing tsk->mm or the
+	 * rq->membarrier_state.
+	 */
+	smp_mb__after_spinlock();
+	local_irq_disable();
 	current->mm = NULL;
-	mmap_read_unlock(mm);
+	membarrier_update_current_mm(NULL);
 	enter_lazy_tlb(mm, current);
+	local_irq_enable();
 	task_unlock(current);
+	mmap_read_unlock(mm);
 	mm_update_next_owner(mm);
 	mmput(mm);
 	if (test_thread_flag(TIF_MEMDIE))
diff --git a/kernel/sched/membarrier.c b/kernel/sched/membarrier.c
index e23e74d..aac3292 100644
--- a/kernel/sched/membarrier.c
+++ b/kernel/sched/membarrier.c
@@ -76,6 +76,18 @@ void membarrier_exec_mmap(struct mm_struct *mm)
 	this_cpu_write(runqueues.membarrier_state, 0);
 }
 
+void membarrier_update_current_mm(struct mm_struct *next_mm)
+{
+	struct rq *rq = this_rq();
+	int membarrier_state = 0;
+
+	if (next_mm)
+		membarrier_state = atomic_read(&next_mm->membarrier_state);
+	if (READ_ONCE(rq->membarrier_state) == membarrier_state)
+		return;
+	WRITE_ONCE(rq->membarrier_state, membarrier_state);
+}
+
 static int membarrier_global_expedited(void)
 {
 	int cpu;