futex: Implement mechanism to wait on any of several futexes

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.

Changes since RFC:
  - Limit waitlist to 128 futexes
  - Simplify wait loop
  - Document functions
  - Reduce allocated space
  - Return hint if a futex was awoken during setup
  - Check if any futex was awoken prior to sleep
  - Drop relative timer logic
Signed-off-by: default avatarZebediah Figura <z.figura12@gmail.com>
Signed-off-by: default avatarSteven Noonan <steven@valvesoftware.com>
Signed-off-by: Pierre-Loup Griffais's avatarPierre-Loup A. Griffais <pgriffais@valvesoftware.com>
Signed-off-by: default avatarGabriel Krisman Bertazi <krisman@collabora.com>
parent ec79671d
......@@ -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 13
#define FUTEX_PRIVATE_FLAG 128
#define FUTEX_CLOCK_REALTIME 256
......@@ -150,4 +151,12 @@ struct robust_list_head {
(((op & 0xf) << 28) | ((cmp & 0xf) << 24) \
| ((oparg & 0xfff) << 12) | (cmparg & 0xfff))
#define FUTEX_MULTIPLE_MAX_COUNT 128
struct futex_wait_block {
__u32 __user *uaddr;
__u32 val;
__u32 bitset;
};
#endif /* _UAPI_LINUX_FUTEX_H */
......@@ -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:
......@@ -237,6 +238,8 @@ struct futex_q {
struct rt_mutex_waiter *rt_waiter;
union futex_key *requeue_pi_key;
u32 bitset;
u32 __user *uaddr;
u32 uval;
} __randomize_layout;
static const struct futex_q futex_q_init = {
......@@ -2344,6 +2347,29 @@ static int unqueue_me(struct futex_q *q)
return ret;
}
/**
* unqueue_multiple() - Remove several futexes from their futex_hash_bucket
* @q: The list of futexes to unqueue
* @count: Number of futexes in the list
*
* Helper to unqueue a list of futexes. This can't fail.
*
* Return:
* - >=0 - Index in the of the last futex that was awoken;
* - -1 - if no futex was awoken
*/
static int unqueue_multiple(struct futex_q *q, int count)
{
int ret = -1;
int i;
for (i = 0; i < count; i++) {
if (!unqueue_me(&q[i]))
ret = i;
}
return ret;
}
/*
* PI futexes can not be requeued and must remove themself from the
* hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
......@@ -2631,22 +2657,21 @@ 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)
static int __futex_wait_setup(struct futex_q *q, unsigned int flags,
struct futex_hash_bucket **hb)
{
u32 uval;
int ret;
retry_private:
*hb = queue_lock(q);
ret = get_futex_value_locked(&uval, uaddr);
ret = get_futex_value_locked(&uval, q->uaddr);
if (ret) {
queue_unlock(*hb);
ret = get_user(uval, uaddr);
ret = get_user(uval, q->uaddr);
if (ret)
return ret;
......@@ -2656,7 +2681,7 @@ static int __futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
return 1;
}
if (uval != val) {
if (uval != q->uval) {
queue_unlock(*hb);
ret = -EWOULDBLOCK;
}
......@@ -2710,7 +2735,9 @@ static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
if (unlikely(ret != 0))
return ret;
ret = __futex_wait_setup(uaddr, val, flags, q, hb);
q->uaddr = uaddr;
q->uval = val;
ret = __futex_wait_setup(q, flags, hb);
/* Drop key reference if retry or error. */
if (ret)
......@@ -2720,6 +2747,176 @@ static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
return ret;
}
/**
* futex_setup_queue_multiple() - Prepare to wait on and enqueue multiple futexes
* @wb: The userspace list of objects to wait on
* @qs: The corresponding futex list
* @count: The size of the lists
* @flags futex flags (FLAGS_SHARED, etc.)
*
* This is a helper to enqueue multiple futex objects in a single step.
* Enqueing multiple futexes are tricky, because we need to enqueue each
* futex in the list before dealing with the next one to avoid
* deadlocking on the hash bucket. In addition, before enqueing, we
* need to make sure that current->state is TASK_INTERRUPTIBLE, but this
* cannot be done before the get_futex_key of the next key, because it
* calls get_user_pages, which can sleep. Thus, we fetch the list of
* futexes keys in two steps, by first pinning all the memory keys in
* the futex key, and only then we read each the key and queue the
* correspoding futex.
*
* Return:
* - 1 - One of the futexes was awaken by another thread
* - 0 - Success
* - <1 - -EFAULT or -EWOULDBLOCK
*/
static int futex_setup_queue_multiple(struct futex_q *qs, int count,
unsigned int flags, int *awaken)
{
struct futex_hash_bucket *hb;
int ret, i;
retry:
for (i = 0; i < count; i++) {
qs[i].key = FUTEX_KEY_INIT;
ret = get_futex_key(qs[i].uaddr, flags & FLAGS_SHARED,
&qs[i].key, FUTEX_READ);
if (unlikely(ret)) {
for (--i; i >= 0; i--)
put_futex_key(&qs[i].key);
return ret;
}
}
set_current_state(TASK_INTERRUPTIBLE);
for (i = 0; i < count; i++) {
ret = __futex_wait_setup(&qs[i], flags, &hb);
if (ret) {
/*
* Keys 0..(i-1) are implicitly put afterwards
* on unqueue_multiple.
*/
put_futex_key(&qs[i].key);
/*
* If something was already awaken, we can
* safely ignore the error and succeed.
*/
*awaken = unqueue_multiple(qs, i);
__set_current_state(TASK_RUNNING);
if (*awaken >= 0)
return 1;
if (ret > 0)
goto retry;
return ret;
}
/*
* The bucket lock can't be held while dealing with the
* next futex. Queue each futex at this moment so hb can
* be unlocked.
*/
queue_me(&qs[i], hb);
}
return 0;
}
/**
* futex_wait_multiple() - Prepare to wait on and enqueue several futexes
* @uaddr: The userspace list of objects to wait on
* @flags: Futex flags (FLAGS_SHARED, etc.)
* @count: The number of objects
* @abs_time: Timeout before giving up and returning to userspace
*
* Entry point for the FUTEX_WAIT_MULTIPLE futex operation, this fuction
* sleeps on a group of futexes and returns on the first futex that
* triggered, or after the timeout has elapsed.
*
* Return:
* - >=0 - Hint to the futex that was awoken
* - <0 - On error
*/
static int futex_wait_multiple(u32 __user *uaddr, unsigned int flags,
u32 count, ktime_t *abs_time)
{
struct hrtimer_sleeper timeout, *to;
struct futex_q *qs = NULL;
int ret, i, hint = 0;
struct futex_wait_block entry, *fwb;
if (!count || count > FUTEX_MULTIPLE_MAX_COUNT)
return -EINVAL;
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);
fwb = (struct futex_wait_block *) __user uaddr;
for (i = 0; i < count; i++) {
if(copy_from_user(&entry, &fwb[i], sizeof(entry)))
return -EFAULT;
qs[i].uval = entry.val;
qs[i].uaddr = entry.uaddr;
qs[i].bitset = entry.bitset;
}
do {
ret = futex_setup_queue_multiple(qs, count, flags, &hint);
if (ret) {
/* A futex was awaken during setup */
if (ret > 0)
ret = hint;
break;
}
if (to)
hrtimer_start_expires(&to->timer, HRTIMER_MODE_ABS);
/*
* Avoid sleeping if another thread already tried to
* wake us.
*/
for (i = 0; i < count; i++) {
if (plist_node_empty(&qs[i].list))
break;
}
if (i == count && (!to || to->task))
freezable_schedule();
ret = unqueue_multiple(qs, count);
__set_current_state(TASK_RUNNING);
if (ret < 0) {
if (to && !to->task) {
ret = -ETIMEDOUT;
break;
} else if (signal_pending(current)) {
ret = -ERESTARTSYS;
break;
}
/*
* The final case is a spurious wakeup, for
* which we can just retry.
*/
}
} while (ret < 0);
if (to) {
hrtimer_cancel(&to->timer);
destroy_hrtimer_on_stack(&to->timer);
}
kfree(qs);
return ret;
}
static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
ktime_t *abs_time, u32 bitset)
{
......@@ -2797,6 +2994,10 @@ 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 +3881,8 @@ 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 +3899,8 @@ 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 +3909,7 @@ 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 +4093,15 @@ 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;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment