Re: [PATCH v3 7/7] locking/rtmutex: Acquire the hb lock via trylock after wait-proxylock.

From: Jiri Slaby
Date: Mon Jan 15 2024 - 06:40:23 EST


On 15. 09. 23, 17:19, Peter Zijlstra wrote:
On Fri, Sep 15, 2023 at 02:58:35PM +0200, Thomas Gleixner wrote:

I spent quite some time to convince myself that this is correct. I was
not able to poke a hole into it. So that really should be safe to
do. Famous last words ...

IKR :-/

Something like so then...

---
Subject: futex/pi: Fix recursive rt_mutex waiter state

So this breaks some random test in APR:

From https://build.opensuse.org/package/live_build_log/openSUSE:Factory:Staging:G/apr/standard/x86_64:
testprocmutex : Line 122: child did not terminate with success

The child in fact terminates on https://github.com/apache/apr/blob/trunk/test/testprocmutex.c#L93:
while ((rv = apr_proc_mutex_timedlock(proc_lock, 1))) {
if (!APR_STATUS_IS_TIMEUP(rv))
exit(1); <----- here

The test creates 6 children and does some pthread_mutex_timedlock/unlock() repeatedly (200 times) in parallel while sleeping 1 us inside the lock. The timeout is 1 us above. And the test expects all them to fail (to time out). But the time out does not always happen in 6.7 (it's racy, so the failure is semi-random: like 1 of 1000 attempts is bad).

If I revert this patch (commit fbeb558b0dd0d), the test works.

I know, the test could be broken too, but I have no idea, really. The testsuite is sort of hairy and I could not come up with a simple repro.

Note APR sets up PTHREAD_PROCESS_SHARED, _ROBUST, and _PRIO_INHERIT attrs for the mutex.

Anyway:
downstream report: https://bugzilla.suse.com/show_bug.cgi?id=1218801
APR report: https://bz.apache.org/bugzilla/show_bug.cgi?id=68481

Any idea if this patch should cause the above (or even is a desired behavior)?

Thanks.

From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Tue, 12 Sep 2023 13:17:11 +0200

Some new assertions pointed out that the existing code has nested rt_mutex wait
state in the futex code.

Specifically, the futex_lock_pi() cancel case uses spin_lock() while there
still is a rt_waiter enqueued for this task, resulting in a state where there
are two waiters for the same task (and task_struct::pi_blocked_on gets
scrambled).

The reason to take hb->lock at this point is to avoid the wake_futex_pi()
EAGAIN case.

This happens when futex_top_waiter() and rt_mutex_top_waiter() state becomes
inconsistent. The current rules are such that this inconsistency will not be
observed.

Notably the case that needs to be avoided is where futex_lock_pi() and
futex_unlock_pi() interleave such that unlock will fail to observe a new
waiter.

*However* the case at hand is where a waiter is leaving, in this case the race
means a waiter that is going away is not observed -- which is harmless,
provided this race is explicitly handled.

This is a somewhat dangerous proposition because the converse race is not
observing a new waiter, which must absolutely not happen. But since the race is
valid this cannot be asserted.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
pi.c | 76 +++++++++++++++++++++++++++++++++++++++-----------------------
requeue.c | 6 +++-
2 files changed, 52 insertions(+), 30 deletions(-)

Index: linux-2.6/kernel/futex/pi.c
===================================================================
--- linux-2.6.orig/kernel/futex/pi.c
+++ linux-2.6/kernel/futex/pi.c
@@ -610,29 +610,16 @@ int futex_lock_pi_atomic(u32 __user *uad
/*
* Caller must hold a reference on @pi_state.
*/
-static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state)
+static int wake_futex_pi(u32 __user *uaddr, u32 uval,
+ struct futex_pi_state *pi_state,
+ struct rt_mutex_waiter *top_waiter)
{
- struct rt_mutex_waiter *top_waiter;
struct task_struct *new_owner;
bool postunlock = false;
DEFINE_RT_WAKE_Q(wqh);
u32 curval, newval;
int ret = 0;
- top_waiter = rt_mutex_top_waiter(&pi_state->pi_mutex);
- if (WARN_ON_ONCE(!top_waiter)) {
- /*
- * As per the comment in futex_unlock_pi() this should not happen.
- *
- * When this happens, give up our locks and try again, giving
- * the futex_lock_pi() instance time to complete, either by
- * waiting on the rtmutex or removing itself from the futex
- * queue.
- */
- ret = -EAGAIN;
- goto out_unlock;
- }
-
new_owner = top_waiter->task;
/*
@@ -1039,19 +1026,33 @@ retry_private:
ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);
cleanup:
- spin_lock(q.lock_ptr);
/*
* If we failed to acquire the lock (deadlock/signal/timeout), we must
- * first acquire the hb->lock before removing the lock from the
- * rt_mutex waitqueue, such that we can keep the hb and rt_mutex wait
- * lists consistent.
+ * must unwind the above, however we canont lock hb->lock because
+ * rt_mutex already has a waiter enqueued and hb->lock can itself try
+ * and enqueue an rt_waiter through rtlock.
+ *
+ * Doing the cleanup without holding hb->lock can cause inconsistent
+ * state between hb and pi_state, but only in the direction of not
+ * seeing a waiter that is leaving.
+ *
+ * See futex_unlock_pi(), it deals with this inconsistency.
+ *
+ * There be dragons here, since we must deal with the inconsistency on
+ * the way out (here), it is impossible to detect/warn about the race
+ * the other way around (missing an incoming waiter).
*
- * In particular; it is important that futex_unlock_pi() can not
- * observe this inconsistency.
+ * What could possibly go wrong...
*/
if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter))
ret = 0;
+ /*
+ * Now that the rt_waiter has been dequeued, it is safe to use
+ * spinlock/rtlock (which might enqueue its own rt_waiter) and fix up
+ * the
+ */
+ spin_lock(q.lock_ptr);
no_block:
/*
* Fixup the pi_state owner and possibly acquire the lock if we
@@ -1132,6 +1133,7 @@ retry:
top_waiter = futex_top_waiter(hb, &key);
if (top_waiter) {
struct futex_pi_state *pi_state = top_waiter->pi_state;
+ struct rt_mutex_waiter *rt_waiter;
ret = -EINVAL;
if (!pi_state)
@@ -1144,22 +1146,39 @@ retry:
if (pi_state->owner != current)
goto out_unlock;
- get_pi_state(pi_state);
/*
* By taking wait_lock while still holding hb->lock, we ensure
- * there is no point where we hold neither; and therefore
- * wake_futex_p() must observe a state consistent with what we
- * observed.
+ * there is no point where we hold neither; and thereby
+ * wake_futex_pi() must observe any new waiters.
+ *
+ * Since the cleanup: case in futex_lock_pi() removes the
+ * rt_waiter without holding hb->lock, it is possible for
+ * wake_futex_pi() to not find a waiter while the above does,
+ * in this case the waiter is on the way out and it can be
+ * ignored.
*
* In particular; this forces __rt_mutex_start_proxy() to
* complete such that we're guaranteed to observe the
- * rt_waiter. Also see the WARN in wake_futex_pi().
+ * rt_waiter.
*/
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
+
+ /*
+ * Futex vs rt_mutex waiter state -- if there are no rt_mutex
+ * waiters even though futex thinks there are, then the waiter
+ * is leaving and the uncontended path is safe to take.
+ */
+ rt_waiter = rt_mutex_top_waiter(&pi_state->pi_mutex);
+ if (!rt_waiter) {
+ raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
+ goto do_uncontended;
+ }
+
+ get_pi_state(pi_state);
spin_unlock(&hb->lock);
/* drops pi_state->pi_mutex.wait_lock */
- ret = wake_futex_pi(uaddr, uval, pi_state);
+ ret = wake_futex_pi(uaddr, uval, pi_state, rt_waiter);
put_pi_state(pi_state);
@@ -1187,6 +1206,7 @@ retry:
return ret;
}
+do_uncontended:
/*
* We have no kernel internal state, i.e. no waiters in the
* kernel. Waiters which are about to queue themselves are stuck
Index: linux-2.6/kernel/futex/requeue.c
===================================================================
--- linux-2.6.orig/kernel/futex/requeue.c
+++ linux-2.6/kernel/futex/requeue.c
@@ -850,11 +850,13 @@ int futex_wait_requeue_pi(u32 __user *ua
pi_mutex = &q.pi_state->pi_mutex;
ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter);
- /* Current is not longer pi_blocked_on */
- spin_lock(q.lock_ptr);
+ /*
+ * See futex_unlock_pi()'s cleanup: comment.
+ */
if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter))
ret = 0;
+ spin_lock(q.lock_ptr);
debug_rt_mutex_free_waiter(&rt_waiter);
/*
* Fixup the pi_state owner and possibly acquire the lock if we


--
js
suse labs