From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-9.7 required=3.0 tests=HEADER_FROM_DIFFERENT_DOMAINS, INCLUDES_PATCH,MAILING_LIST_MULTI,SIGNED_OFF_BY,SPF_HELO_NONE,SPF_PASS, UNPARSEABLE_RELAY,URIBL_BLOCKED,USER_AGENT_GIT autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 1B862C433FF for ; Tue, 30 Jul 2019 22:06:55 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id DFC3C206A2 for ; Tue, 30 Jul 2019 22:06:54 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S2388012AbfG3WGx (ORCPT ); Tue, 30 Jul 2019 18:06:53 -0400 Received: from bhuna.collabora.co.uk ([46.235.227.227]:44510 "EHLO bhuna.collabora.co.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S2387994AbfG3WGw (ORCPT ); Tue, 30 Jul 2019 18:06:52 -0400 Received: from [127.0.0.1] (localhost [127.0.0.1]) (Authenticated sender: krisman) with ESMTPSA id DA02128B70F From: Gabriel Krisman Bertazi To: tglx@linutronix.de Cc: mingo@redhat.com, peterz@infradead.org, dvhart@infradead.org, linux-kernel@vger.kernel.org, Gabriel Krisman Bertazi , kernel@collabora.com, Zebediah Figura , Steven Noonan , "Pierre-Loup A . Griffais" Subject: [PATCH RFC 2/2] futex: Implement mechanism to wait on any of several futexes Date: Tue, 30 Jul 2019 18:06:02 -0400 Message-Id: <20190730220602.28781-2-krisman@collabora.com> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20190730220602.28781-1-krisman@collabora.com> References: <20190730220602.28781-1-krisman@collabora.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Sender: linux-kernel-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-kernel@vger.kernel.org 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 Signed-off-by: Steven Noonan Signed-off-by: Pierre-Loup A. Griffais Signed-off-by: Gabriel Krisman Bertazi --- 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 13 #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 = ¤t->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