Skip to contentSkip to conversation

locking/rwsem: Add missing ACQUIRE to read_slowpath exit when queue is empty

PATCH In review Sasha Levin 2 participants 4 messages Aug 26, 2019 – Aug 29, 2019

Subsystems (2)
lore
Share & tools
lore

Sasha Levin

author maintainer
#0 lore raw

PATCH 1/2 locking/rwsem: Add missing ACQUIRE to read_slowpath exit when queue is empty

From: Jan Stancek <[email protected]>

[ Upstream commit e1b98fa316648420d0434d9ff5b92ad6609ba6c3 ]

LTP mtest06 has been observed to occasionally hit "still mapped when deleted" and following BUG_ON on arm64.

The extra mapcount originated from pagefault handler, which handled pagefault for vma that has already been detached. vma is detached under mmap_sem write lock by detach_vmas_to_be_unmapped(), which also invalidates vmacache.

When the pagefault handler (under mmap_sem read lock) calls find_vma(), vmacache_valid() wrongly reports vmacache as valid.

After rwsem down_read() returns via 'queue empty' path (as of v5.2), it does so without an ACQUIRE on sem->count:

  down_read()
    __down_read()
      rwsem_down_read_failed()
        __rwsem_down_read_failed_common()
          raw_spin_lock_irq(&sem->wait_lock);
          if (list_empty(&sem->wait_list)) {
            if (atomic_long_read(&sem->count) >= 0) {
              raw_spin_unlock_irq(&sem->wait_lock);
              return sem;

The problem can be reproduced by running LTP mtest06 in a loop and building the kernel (-j $NCPUS) in parallel. It does reproduces since v4.20 on arm64 HPE Apollo 70 (224 CPUs, 256GB RAM, 2 nodes). It triggers reliably in about an hour.

The patched kernel ran fine for 10+ hours.

---

This is a backport for the v5.2 stable tree. There were multiple reports of this issue being hit.

Given that there were a few changes to the code around this, I'd appreciate an ack before pulling it in.

 kernel/locking/rwsem-xadd.c | 2 ++
 1 file changed, 2 insertions(+)
kernel/locking/rwsem-xadd.c +2 -0
index 0b1f779572402..397dedc58432d 100644
@@ -454,6 +454,8 @@ __rwsem_down_read_failed_common(struct rw_semaphore *sem, int state)
 		 * been set in the count.
 		 */
 		if (atomic_long_read(&sem->count) >= 0) {
+			/* Provide lock ACQUIRE */
+			smp_acquire__after_ctrl_dep();
 			raw_spin_unlock_irq(&sem->wait_lock);
 			rwsem_set_reader_owned(sem);
 			lockevent_inc(rwsem_rlock_fast);
Signature · 1 line
2.20.1
author maintainer
#1 lore raw

PATCH 2/2 lcoking/rwsem: Add missing ACQUIRE to read_slowpath sleep loop

From: Peter Zijlstra <[email protected]>

[ Upstream commit 99143f82a255e7f054bead8443462fae76dd829e ]

While reviewing another read_slowpath patch, both Will and I noticed another missing ACQUIRE, namely:

  X = 0;

CPU0 CPU1

  rwsem_down_read()
    for (;;) {
      set_current_state(TASK_UNINTERRUPTIBLE);
                        X = 1;
                        rwsem_up_write();
                          rwsem_mark_wake()
                            atomic_long_add(adjustment, &sem->count);
                            smp_store_release(&waiter->task, NULL);

      if (!waiter.task)
        break;

      ...
    }
  r = X;

Allows 'r == 0'.

---

This is a backport for the v5.2 stable tree. There were multiple reports of this issue being hit.

Given that there were a few changes to the code around this, I'd appreciate an ack before pulling it in.

 kernel/locking/rwsem-xadd.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
kernel/locking/rwsem-xadd.c +3 -1
index 397dedc58432d..385ebcfc31a6d 100644
@@ -485,8 +485,10 @@ __rwsem_down_read_failed_common(struct rw_semaphore *sem, int state)
 	/* wait to be given the lock */
 	while (true) {
 		set_current_state(state);
-		if (!waiter.task)
+		if (!smp_load_acquire(&waiter.task)) {
+			/* Orders against rwsem_mark_wake()'s smp_store_release() */
 			break;
+		}
 		if (signal_pending_state(state, current)) {
 			raw_spin_lock_irq(&sem->wait_lock);
 			if (waiter.task)
Signature · 1 line
2.20.1
#2 lore raw
In reply to Sasha Levin

----- Original Message -----

Quoted 49 lines

From: Jan Stancek <[email protected]>

[ Upstream commit e1b98fa316648420d0434d9ff5b92ad6609ba6c3 ]

LTP mtest06 has been observed to occasionally hit "still mapped when deleted" and following BUG_ON on arm64.

The extra mapcount originated from pagefault handler, which handled pagefault for vma that has already been detached. vma is detached under mmap_sem write lock by detach_vmas_to_be_unmapped(), which also invalidates vmacache.

When the pagefault handler (under mmap_sem read lock) calls find_vma(), vmacache_valid() wrongly reports vmacache as valid.

After rwsem down_read() returns via 'queue empty' path (as of v5.2), it does so without an ACQUIRE on sem->count:

  down_read()
    __down_read()
      rwsem_down_read_failed()
        __rwsem_down_read_failed_common()
          raw_spin_lock_irq(&sem->wait_lock);
          if (list_empty(&sem->wait_list)) {
            if (atomic_long_read(&sem->count) >= 0) {
              raw_spin_unlock_irq(&sem->wait_lock);
              return sem;

The problem can be reproduced by running LTP mtest06 in a loop and building the kernel (-j $NCPUS) in parallel. It does reproduces since v4.20 on arm64 HPE Apollo 70 (224 CPUs, 256GB RAM, 2 nodes). It triggers reliably in about an hour.

The patched kernel ran fine for 10+ hours.

Cc Linus Torvalds <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Fixes 4b486b535c33 ("locking/rwsem: Exit read lock slowpath if queue empty &

no writer")
Link:
https://lkml.kernel.org/r/50b8914e20d1d62bb2dee42d342836c2c16ebee7.1563438048.git.jstancek@redhat.com

Quoted 7 lines

---

This is a backport for the v5.2 stable tree. There were multiple reports of this issue being hit.

Given that there were a few changes to the code around this, I'd appreciate an ack before pulling it in.

ACK, both look good to me.
I also re-ran reproducer with this series applied on top of 5.2.10, it PASS-ed.

Thanks,
Jan

author maintainer Applied
#3 lore raw
Quoted 2 lines · Jan Stancek

ACK, both look good to me.
I also re-ran reproducer with this series applied on top of 5.2.10, it PASS-ed.

I've queued both for 5.2, thanks Jan.

--
Thanks,
Sasha