path: root/0007-v5.4-fsync.patch

split the futex key setup from the queue locking and key reading.  This
is useful to support the setup of multiple keys at the same time, like
what is done in futex_requeue() and what will be done for the
FUTEX_WAIT_MULTIPLE command.

Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
---
 kernel/futex.c | 71 +++++++++++++++++++++++++++++---------------------
 1 file changed, 42 insertions(+), 29 deletions(-)

diff --git a/kernel/futex.c b/kernel/futex.c
index 6d50728ef2e7..91f3db335c57 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -2631,6 +2631,39 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
 	__set_current_state(TASK_RUNNING);
 }
 
+static int __futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
+			      struct futex_q *q, struct futex_hash_bucket **hb)
+{
+
+	u32 uval;
+	int ret;
+
+retry_private:
+	*hb = queue_lock(q);
+
+	ret = get_futex_value_locked(&uval, uaddr);
+
+	if (ret) {
+		queue_unlock(*hb);
+
+		ret = get_user(uval, uaddr);
+		if (ret)
+			return ret;
+
+		if (!(flags & FLAGS_SHARED))
+			goto retry_private;
+
+		return 1;
+	}
+
+	if (uval != val) {
+		queue_unlock(*hb);
+		ret = -EWOULDBLOCK;
+	}
+
+	return ret;
+}
+
 /**
  * futex_wait_setup() - Prepare to wait on a futex
  * @uaddr:	the futex userspace address
@@ -2651,7 +2684,6 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
 static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
 			   struct futex_q *q, struct futex_hash_bucket **hb)
 {
-	u32 uval;
 	int ret;
 
 	/*
@@ -2672,38 +2704,19 @@ static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
 	 * absorb a wakeup if *uaddr does not match the desired values
 	 * while the syscall executes.
 	 */
-retry:
-	ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, FUTEX_READ);
-	if (unlikely(ret != 0))
-		return ret;
-
-retry_private:
-	*hb = queue_lock(q);
+	do {
+		ret = get_futex_key(uaddr, flags & FLAGS_SHARED,
+				    &q->key, FUTEX_READ);
+		if (unlikely(ret != 0))
+			return ret;
 
-	ret = get_futex_value_locked(&uval, uaddr);
+		ret = __futex_wait_setup(uaddr, val, flags, q, hb);
 
-	if (ret) {
-		queue_unlock(*hb);
-
-		ret = get_user(uval, uaddr);
+		/* Drop key reference if retry or error. */
 		if (ret)
-			goto out;
+			put_futex_key(&q->key);
+	} while (ret > 0);
 
-		if (!(flags & FLAGS_SHARED))
-			goto retry_private;
-
-		put_futex_key(&q->key);
-		goto retry;
-	}
-
-	if (uval != val) {
-		queue_unlock(*hb);
-		ret = -EWOULDBLOCK;
-	}
-
-out:
-	if (ret)
-		put_futex_key(&q->key);
 	return ret;
 }
 
-- 
2.20.1 

This is a new futex operation, called FUTEX_WAIT_MULTIPLE, which allows
a thread to wait on several futexes at the same time, and be awoken by
any of them.  In a sense, it implements one of the features that was
supported by pooling on the old FUTEX_FD interface.

My use case for this operation lies in Wine, where we want to implement
a similar interface available in Windows, used mainly for event
handling.  The wine folks have an implementation that uses eventfd, but
it suffers from FD exhaustion (I was told they have application that go
to the order of multi-milion FDs), and higher CPU utilization.

In time, we are also proposing modifications to glibc and libpthread to
make this feature available for Linux native multithreaded applications
using libpthread, which can benefit from the behavior of waiting on any
of a group of futexes.

In particular, using futexes in our Wine use case reduced the CPU
utilization by 4% for the game Beat Saber and by 1.5% for the game
Shadow of Tomb Raider, both running over Proton (a wine based solution
for Windows emulation), when compared to the eventfd interface. This
implementation also doesn't rely of file descriptors, so it doesn't risk
overflowing the resource.

Technically, the existing FUTEX_WAIT implementation can be easily
reworked by using do_futex_wait_multiple with a count of one, and I
have a patch showing how it works.  I'm not proposing it, since
futex is such a tricky code, that I'd be more confortable to have
FUTEX_WAIT_MULTIPLE running upstream for a couple development cycles,
before considering modifying FUTEX_WAIT.

From an implementation perspective, the futex list is passed as an array
of (pointer,value,bitset) to the kernel, which will enqueue all of them
and sleep if none was already triggered. It returns a hint of which
futex caused the wake up event to userspace, but the hint doesn't
guarantee that is the only futex triggered.  Before calling the syscall
again, userspace should traverse the list, trying to re-acquire any of
the other futexes, to prevent an immediate -EWOULDBLOCK return code from
the kernel.

This was tested using three mechanisms:

1) By reimplementing FUTEX_WAIT in terms of FUTEX_WAIT_MULTIPLE and
running the unmodified tools/testing/selftests/futex and a full linux
distro on top of this kernel.

2) By an example code that exercises the FUTEX_WAIT_MULTIPLE path on a
multi-threaded, event-handling setup.

3) By running the Wine fsync implementation and executing multi-threaded
applications, in particular the modern games mentioned above, on top of
this implementation.

Signed-off-by: Zebediah Figura <z.figura12@gmail.com>
Signed-off-by: Steven Noonan <steven@valvesoftware.com>
Signed-off-by: Pierre-Loup A. Griffais <pgriffais@valvesoftware.com>
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
---
 include/uapi/linux/futex.h |   7 ++
 kernel/futex.c             | 161 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 164 insertions(+), 4 deletions(-)

diff --git a/include/uapi/linux/futex.h b/include/uapi/linux/futex.h
index a89eb0accd5e..2401c4cf5095 100644
--- a/include/uapi/linux/futex.h
+++ b/include/uapi/linux/futex.h
@@ -21,6 +21,7 @@
 #define FUTEX_WAKE_BITSET	10
 #define FUTEX_WAIT_REQUEUE_PI	11
 #define FUTEX_CMP_REQUEUE_PI	12
+#define FUTEX_WAIT_MULTIPLE	31
 
 #define FUTEX_PRIVATE_FLAG	128
 #define FUTEX_CLOCK_REALTIME	256
@@ -150,4 +151,10 @@ struct robust_list_head {
   (((op & 0xf) << 28) | ((cmp & 0xf) << 24)		\
    | ((oparg & 0xfff) << 12) | (cmparg & 0xfff))
 
+struct futex_wait_block {
+	__u32 __user *uaddr;
+	__u32 val;
+	__u32 bitset;
+};
+
 #endif /* _UAPI_LINUX_FUTEX_H */
diff --git a/kernel/futex.c b/kernel/futex.c
index 91f3db335c57..2623e8f152cd 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -183,6 +183,7 @@ static int  __read_mostly futex_cmpxchg_enabled;
 #endif
 #define FLAGS_CLOCKRT		0x02
 #define FLAGS_HAS_TIMEOUT	0x04
+#define FLAGS_WAKE_MULTIPLE	0x08
 
 /*
  * Priority Inheritance state:
@@ -2720,6 +2721,150 @@ static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
 	return ret;
 }
 
+static int do_futex_wait_multiple(struct futex_wait_block *wb,
+				  u32 count, unsigned int flags,
+				  ktime_t *abs_time)
+{
+
+	struct hrtimer_sleeper timeout, *to;
+	struct futex_hash_bucket *hb;
+	struct futex_q *qs = NULL;
+	int ret;
+	int i;
+
+	qs = kcalloc(count, sizeof(struct futex_q), GFP_KERNEL);
+	if (!qs)
+		return -ENOMEM;
+
+	to = futex_setup_timer(abs_time, &timeout, flags,
+			       current->timer_slack_ns);
+ retry:
+	for (i = 0; i < count; i++) {
+		qs[i].key = FUTEX_KEY_INIT;
+		qs[i].bitset = wb[i].bitset;
+
+		ret = get_futex_key(wb[i].uaddr, flags & FLAGS_SHARED,
+				    &qs[i].key, FUTEX_READ);
+		if (unlikely(ret != 0)) {
+			for (--i; i >= 0; i--)
+				put_futex_key(&qs[i].key);
+			goto out;
+		}
+	}
+
+	set_current_state(TASK_INTERRUPTIBLE);
+
+	for (i = 0; i < count; i++) {
+		ret = __futex_wait_setup(wb[i].uaddr, wb[i].val,
+					 flags, &qs[i], &hb);
+		if (ret) {
+			/* Drop the failed key directly.  keys 0..(i-1)
+			 * will be put by unqueue_me.
+			 */
+			put_futex_key(&qs[i].key);
+
+			/* Undo the partial work we did. */
+			for (--i; i >= 0; i--)
+				unqueue_me(&qs[i]);
+
+			__set_current_state(TASK_RUNNING);
+			if (ret > 0)
+				goto retry;
+			goto out;
+		}
+
+		/* We can't hold to the bucket lock when dealing with
+		 * the next futex. Queue ourselves now so we can unlock
+		 * it before moving on.
+		 */
+		queue_me(&qs[i], hb);
+	}
+
+	if (to)
+		hrtimer_start_expires(&to->timer, HRTIMER_MODE_ABS);
+
+	/* There is no easy to way to check if we are wake already on
+	 * multiple futexes without waking through each one of them.  So
+	 * just sleep and let the scheduler handle it.
+	 */
+	if (!to || to->task)
+		freezable_schedule();
+
+	__set_current_state(TASK_RUNNING);
+
+	ret = -ETIMEDOUT;
+	/* If we were woken (and unqueued), we succeeded. */
+	for (i = 0; i < count; i++)
+		if (!unqueue_me(&qs[i]))
+			ret = i;
+
+	/* Succeed wakeup */
+	if (ret >= 0)
+		goto out;
+
+	/* Woken by triggered timeout */
+	if (to && !to->task)
+		goto out;
+
+	/*
+	 * We expect signal_pending(current), but we might be the
+	 * victim of a spurious wakeup as well.
+	 */
+	if (!signal_pending(current))
+		goto retry;
+
+	ret = -ERESTARTSYS;
+	if (!abs_time)
+		goto out;
+
+	ret = -ERESTART_RESTARTBLOCK;
+ out:
+	if (to) {
+		hrtimer_cancel(&to->timer);
+		destroy_hrtimer_on_stack(&to->timer);
+	}
+
+	kfree(qs);
+	return ret;
+}
+
+static int futex_wait_multiple(u32 __user *uaddr, unsigned int flags,
+			       u32 count, ktime_t *abs_time)
+{
+	struct futex_wait_block *wb;
+	struct restart_block *restart;
+	int ret;
+
+	if (!count)
+		return -EINVAL;
+
+	wb = kcalloc(count, sizeof(struct futex_wait_block), GFP_KERNEL);
+	if (!wb)
+		return -ENOMEM;
+
+	if (copy_from_user(wb, uaddr,
+			   count * sizeof(struct futex_wait_block))) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	ret = do_futex_wait_multiple(wb, count, flags, abs_time);
+
+	if (ret == -ERESTART_RESTARTBLOCK) {
+		restart = &current->restart_block;
+		restart->fn = futex_wait_restart;
+		restart->futex.uaddr = uaddr;
+		restart->futex.val = count;
+		restart->futex.time = *abs_time;
+		restart->futex.flags = (flags | FLAGS_HAS_TIMEOUT |
+					FLAGS_WAKE_MULTIPLE);
+	}
+
+out:
+	kfree(wb);
+	return ret;
+}
+
 static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
 		      ktime_t *abs_time, u32 bitset)
 {
@@ -2797,6 +2942,10 @@ static long futex_wait_restart(struct restart_block *restart)
 	}
 	restart->fn = do_no_restart_syscall;
 
+	if (restart->futex.flags & FLAGS_WAKE_MULTIPLE)
+		return (long)futex_wait_multiple(uaddr, restart->futex.flags,
+						 restart->futex.val, tp);
+
 	return (long)futex_wait(uaddr, restart->futex.flags,
 				restart->futex.val, tp, restart->futex.bitset);
 }
@@ -3680,6 +3829,8 @@ long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
 					     uaddr2);
 	case FUTEX_CMP_REQUEUE_PI:
 		return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
+	case FUTEX_WAIT_MULTIPLE:
+		return futex_wait_multiple(uaddr, flags, val, timeout);
 	}
 	return -ENOSYS;
 }
@@ -3696,7 +3847,8 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
 
 	if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
 		      cmd == FUTEX_WAIT_BITSET ||
-		      cmd == FUTEX_WAIT_REQUEUE_PI)) {
+		      cmd == FUTEX_WAIT_REQUEUE_PI ||
+		      cmd == FUTEX_WAIT_MULTIPLE)) {
 		if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG))))
 			return -EFAULT;
 		if (get_timespec64(&ts, utime))
@@ -3705,7 +3857,7 @@ SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
 			return -EINVAL;
 
 		t = timespec64_to_ktime(ts);
-		if (cmd == FUTEX_WAIT)
+		if (cmd == FUTEX_WAIT || cmd == FUTEX_WAIT_MULTIPLE)
 			t = ktime_add_safe(ktime_get(), t);
 		tp = &t;
 	}
@@ -3889,14 +4041,15 @@ SYSCALL_DEFINE6(futex_time32, u32 __user *, uaddr, int, op, u32, val,
 
 	if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
 		      cmd == FUTEX_WAIT_BITSET ||
-		      cmd == FUTEX_WAIT_REQUEUE_PI)) {
+		      cmd == FUTEX_WAIT_REQUEUE_PI ||
+		      cmd == FUTEX_WAIT_MULTIPLE)) {
 		if (get_old_timespec32(&ts, utime))
 			return -EFAULT;
 		if (!timespec64_valid(&ts))
 			return -EINVAL;
 
 		t = timespec64_to_ktime(ts);
-		if (cmd == FUTEX_WAIT)
+		if (cmd == FUTEX_WAIT || cmd == FUTEX_WAIT_MULTIPLE)
 			t = ktime_add_safe(ktime_get(), t);
 		tp = &t;
 	}
-- 
2.20.1