From: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To: stable@vger.kernel.org
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
patches@lists.linux.dev, Roman Gershman <romger@amazon.com>,
Jens Axboe <axboe@kernel.dk>,
Thomas Gleixner <tglx@linutronix.de>,
Oleg Nesterov <oleg@redhat.com>
Subject: [PATCH 5.10 16/63] task_work: Use TIF_NOTIFY_SIGNAL if available
Date: Tue, 3 Jan 2023 09:13:46 +0100 [thread overview]
Message-ID: <20230103081309.539024685@linuxfoundation.org> (raw)
In-Reply-To: <20230103081308.548338576@linuxfoundation.org>
From: Jens Axboe <axboe@kernel.dk>
[ Upstream commit 114518eb6430b832d2f9f5a008043b913ccf0e24 ]
If the arch supports TIF_NOTIFY_SIGNAL, then use that for TWA_SIGNAL as
it's more efficient than using the signal delivery method. This is
especially true on threaded applications, where ->sighand is shared across
threads, but it's also lighter weight on non-shared cases.
io_uring is a heavy consumer of TWA_SIGNAL based task_work. A test with
threads shows a nice improvement running an io_uring based echo server.
stock kernel:
0.01% <= 0.1 milliseconds
95.86% <= 0.2 milliseconds
98.27% <= 0.3 milliseconds
99.71% <= 0.4 milliseconds
100.00% <= 0.5 milliseconds
100.00% <= 0.6 milliseconds
100.00% <= 0.7 milliseconds
100.00% <= 0.8 milliseconds
100.00% <= 0.9 milliseconds
100.00% <= 1.0 milliseconds
100.00% <= 1.1 milliseconds
100.00% <= 2 milliseconds
100.00% <= 3 milliseconds
100.00% <= 3 milliseconds
1378930.00 requests per second
~1600% CPU
1.38M requests/second, and all 16 CPUs are maxed out.
patched kernel:
0.01% <= 0.1 milliseconds
98.24% <= 0.2 milliseconds
99.47% <= 0.3 milliseconds
99.99% <= 0.4 milliseconds
100.00% <= 0.5 milliseconds
100.00% <= 0.6 milliseconds
100.00% <= 0.7 milliseconds
100.00% <= 0.8 milliseconds
100.00% <= 0.9 milliseconds
100.00% <= 1.2 milliseconds
1666111.38 requests per second
~1450% CPU
1.67M requests/second, and we're no longer just hammering on the sighand
lock. The original reporter states:
"For 5.7.15 my benchmark achieves 1.6M qps and system cpu is at ~80%.
for 5.7.16 or later it achieves only 1M qps and the system cpu is is
at ~100%"
with the only difference there being that TWA_SIGNAL is used
unconditionally in 5.7.16, since it's required to be able to handle the
inability to run task_work if the application is waiting in the kernel
already on an event that needs task_work run to be satisfied. Also see
commit 0ba9c9edcd15.
Reported-by: Roman Gershman <romger@amazon.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Link: https://lore.kernel.org/r/20201026203230.386348-5-axboe@kernel.dk
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
kernel/task_work.c | 41 +++++++++++++++++++++++++++++------------
1 file changed, 29 insertions(+), 12 deletions(-)
--- a/kernel/task_work.c
+++ b/kernel/task_work.c
@@ -5,6 +5,34 @@
static struct callback_head work_exited; /* all we need is ->next == NULL */
+/*
+ * TWA_SIGNAL signaling - use TIF_NOTIFY_SIGNAL, if available, as it's faster
+ * than TIF_SIGPENDING as there's no dependency on ->sighand. The latter is
+ * shared for threads, and can cause contention on sighand->lock. Even for
+ * the non-threaded case TIF_NOTIFY_SIGNAL is more efficient, as no locking
+ * or IRQ disabling is involved for notification (or running) purposes.
+ */
+static void task_work_notify_signal(struct task_struct *task)
+{
+#if defined(TIF_NOTIFY_SIGNAL)
+ set_notify_signal(task);
+#else
+ unsigned long flags;
+
+ /*
+ * Only grab the sighand lock if we don't already have some
+ * task_work pending. This pairs with the smp_store_mb()
+ * in get_signal(), see comment there.
+ */
+ if (!(READ_ONCE(task->jobctl) & JOBCTL_TASK_WORK) &&
+ lock_task_sighand(task, &flags)) {
+ task->jobctl |= JOBCTL_TASK_WORK;
+ signal_wake_up(task, 0);
+ unlock_task_sighand(task, &flags);
+ }
+#endif
+}
+
/**
* task_work_add - ask the @task to execute @work->func()
* @task: the task which should run the callback
@@ -33,7 +61,6 @@ int task_work_add(struct task_struct *ta
enum task_work_notify_mode notify)
{
struct callback_head *head;
- unsigned long flags;
do {
head = READ_ONCE(task->task_works);
@@ -49,17 +76,7 @@ int task_work_add(struct task_struct *ta
set_notify_resume(task);
break;
case TWA_SIGNAL:
- /*
- * Only grab the sighand lock if we don't already have some
- * task_work pending. This pairs with the smp_store_mb()
- * in get_signal(), see comment there.
- */
- if (!(READ_ONCE(task->jobctl) & JOBCTL_TASK_WORK) &&
- lock_task_sighand(task, &flags)) {
- task->jobctl |= JOBCTL_TASK_WORK;
- signal_wake_up(task, 0);
- unlock_task_sighand(task, &flags);
- }
+ task_work_notify_signal(task);
break;
default:
WARN_ON_ONCE(1);
next prev parent reply other threads:[~2023-01-03 8:16 UTC|newest]
Thread overview: 74+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-01-03 8:13 [PATCH 5.10 00/63] 5.10.162-rc1 review Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 01/63] kernel: provide create_io_thread() helper Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 02/63] iov_iter: add helper to save iov_iter state Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 03/63] saner calling conventions for unlazy_child() Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 04/63] fs: add support for LOOKUP_CACHED Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 05/63] fix handling of nd->depth on LOOKUP_CACHED failures in try_to_unlazy* Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 06/63] Make sure nd->path.mnt and nd->path.dentry are always valid pointers Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 07/63] fs: expose LOOKUP_CACHED through openat2() RESOLVE_CACHED Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 08/63] tools headers UAPI: Sync openat2.h with the kernel sources Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 09/63] net: provide __sys_shutdown_sock() that takes a socket Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 10/63] net: add accept helper not installing fd Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 11/63] signal: Add task_sigpending() helper Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 12/63] fs: make do_renameat2() take struct filename Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 13/63] file: Rename __close_fd_get_file close_fd_get_file Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 14/63] fs: provide locked helper variant of close_fd_get_file() Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 15/63] entry: Add support for TIF_NOTIFY_SIGNAL Greg Kroah-Hartman
2023-01-03 8:13 ` Greg Kroah-Hartman [this message]
2023-01-03 8:13 ` [PATCH 5.10 17/63] x86: Wire up TIF_NOTIFY_SIGNAL Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 18/63] arc: add support for TIF_NOTIFY_SIGNAL Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 19/63] arm64: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 20/63] m68k: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 21/63] nios32: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 22/63] parisc: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 23/63] powerpc: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 24/63] mips: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 25/63] s390: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 26/63] um: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 27/63] sh: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 28/63] openrisc: " Greg Kroah-Hartman
2023-01-03 8:13 ` [PATCH 5.10 29/63] csky: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 30/63] hexagon: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 31/63] microblaze: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 32/63] arm: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 33/63] xtensa: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 34/63] alpha: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 35/63] c6x: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 36/63] h8300: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 37/63] ia64: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 38/63] nds32: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 39/63] riscv: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 40/63] sparc: " Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 41/63] ia64: dont call handle_signal() unless theres actually a signal queued Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 42/63] ARC: unbork 5.11 bootup: fix snafu in _TIF_NOTIFY_SIGNAL handling Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 43/63] alpha: fix TIF_NOTIFY_SIGNAL handling Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 44/63] task_work: remove legacy TWA_SIGNAL path Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 45/63] kernel: remove checking for TIF_NOTIFY_SIGNAL Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 46/63] coredump: Limit what can interrupt coredumps Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 47/63] kernel: allow fork with TIF_NOTIFY_SIGNAL pending Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 48/63] entry/kvm: Exit to user mode when TIF_NOTIFY_SIGNAL is set Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 49/63] arch: setup PF_IO_WORKER threads like PF_KTHREAD Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 50/63] arch: ensure parisc/powerpc handle PF_IO_WORKER in copy_thread() Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 51/63] x86/process: setup io_threads more like normal user space threads Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 52/63] kernel: stop masking signals in create_io_thread() Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 53/63] kernel: dont call do_exit() for PF_IO_WORKER threads Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 54/63] task_work: add helper for more targeted task_work canceling Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 55/63] io_uring: import 5.15-stable io_uring Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 56/63] signal: kill JOBCTL_TASK_WORK Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 57/63] task_work: unconditionally run task_work from get_signal() Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 58/63] net: remove cmsg restriction from io_uring based send/recvmsg calls Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 59/63] Revert "proc: dont allow async path resolution of /proc/thread-self components" Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 60/63] Revert "proc: dont allow async path resolution of /proc/self components" Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 61/63] eventpoll: add EPOLL_URING_WAKE poll wakeup flag Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 62/63] eventfd: provide a eventfd_signal_mask() helper Greg Kroah-Hartman
2023-01-03 8:14 ` [PATCH 5.10 63/63] io_uring: pass in EPOLL_URING_WAKE for eventfd signaling and wakeups Greg Kroah-Hartman
2023-01-03 13:44 ` [PATCH 5.10 00/63] 5.10.162-rc1 review Pavel Machek
2023-01-03 14:51 ` Guenter Roeck
2023-01-03 15:32 ` Jon Hunter
2023-01-03 16:16 ` Joel Fernandes
2023-01-04 5:29 ` Greg Kroah-Hartman
2023-01-04 21:56 ` Joel Fernandes
2023-01-05 11:43 ` Greg Kroah-Hartman
2023-01-03 18:16 ` Naresh Kamboju
2023-01-03 18:27 ` Florian Fainelli
2023-01-03 18:59 ` Allen Pais
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20230103081309.539024685@linuxfoundation.org \
--to=gregkh@linuxfoundation.org \
--cc=axboe@kernel.dk \
--cc=oleg@redhat.com \
--cc=patches@lists.linux.dev \
--cc=romger@amazon.com \
--cc=stable@vger.kernel.org \
--cc=tglx@linutronix.de \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox