mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
From: Stuart Hayes <stuart.w.hayes@gmail.com>
To: linux-kernel@vger.kernel.org,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	"Rafael J . Wysocki" <rafael@kernel.org>,
	Martin Belanger <Martin.Belanger@dell.com>,
	Oliver O'Halloran <oohall@gmail.com>,
	Daniel Wagner <dwagner@suse.de>, Keith Busch <kbusch@kernel.org>,
	Lukas Wunner <lukas@wunner.de>,
	David Jeffery <djeffery@redhat.com>,
	Jeremy Allison <jallison@ciq.com>, Jens Axboe <axboe@fb.com>,
	Christoph Hellwig <hch@lst.de>, Sagi Grimberg <sagi@grimberg.me>,
	linux-nvme@lists.infradead.org,
	Nathan Chancellor <nathan@kernel.org>,
	Jan Kiszka <jan.kiszka@seimens.com>,
	Bert Karwatzki <spasswolf@web.de>
Cc: Stuart Hayes <stuart.w.hayes@gmail.com>
Subject: [PATCH v10 1/5] kernel/async: streamline cookie synchronization
Date: Wed, 25 Jun 2025 15:18:49 -0500	[thread overview]
Message-ID: <20250625201853.84062-2-stuart.w.hayes@gmail.com> (raw)
In-Reply-To: <20250625201853.84062-1-stuart.w.hayes@gmail.com>

From: David Jeffery <djeffery@redhat.com>

To prevent a thundering herd effect, implement a custom wake function for
the async shubsystem which will only wake waiters which have all their
dependencies completed.

The async subsystem currently wakes all waiters on async_done when an async
task completes. When there are many tasks trying to synchronize on differnt
async values, this can create a thundering herd problem when an async task
wakes up all waiters, most of whom go back to waiting after causing
lock contention and wasting CPU.

Signed-off-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Stuart Hayes <stuart.w.hayes@gmail.com>
---
 kernel/async.c | 42 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 41 insertions(+), 1 deletion(-)

diff --git a/kernel/async.c b/kernel/async.c
index 4c3e6a44595f..ae327f29bac9 100644
--- a/kernel/async.c
+++ b/kernel/async.c
@@ -76,6 +76,12 @@ struct async_entry {
 	struct async_domain	*domain;
 };
 
+struct async_wait_entry {
+	wait_queue_entry_t wait;
+	async_cookie_t cookie;
+	struct async_domain *domain;
+};
+
 static DECLARE_WAIT_QUEUE_HEAD(async_done);
 
 static atomic_t entry_count;
@@ -298,6 +304,24 @@ void async_synchronize_full_domain(struct async_domain *domain)
 }
 EXPORT_SYMBOL_GPL(async_synchronize_full_domain);
 
+/**
+ * async_domain_wake_function - wait function for cooking synchronization
+ *
+ * Custom wait function for async_synchronize_cookie_domain to check cookie
+ * value.  This prevents waking up waiting threads unnecessarily.
+ */
+static int async_domain_wake_function(struct wait_queue_entry *wait,
+				      unsigned int mode, int sync, void *key)
+{
+	struct async_wait_entry *await =
+		container_of(wait, struct async_wait_entry, wait);
+
+	if (lowest_in_progress(await->domain) < await->cookie)
+		return 0;
+
+	return autoremove_wake_function(wait, mode, sync, key);
+}
+
 /**
  * async_synchronize_cookie_domain - synchronize asynchronous function calls within a certain domain with cookie checkpointing
  * @cookie: async_cookie_t to use as checkpoint
@@ -310,11 +334,27 @@ EXPORT_SYMBOL_GPL(async_synchronize_full_domain);
 void async_synchronize_cookie_domain(async_cookie_t cookie, struct async_domain *domain)
 {
 	ktime_t starttime;
+	struct async_wait_entry await = {
+		.cookie = cookie,
+		.domain = domain,
+		.wait = {
+			.func = async_domain_wake_function,
+			.private = current,
+			.flags = 0,
+			.entry = LIST_HEAD_INIT(await.wait.entry),
+		}};
 
 	pr_debug("async_waiting @ %i\n", task_pid_nr(current));
 	starttime = ktime_get();
 
-	wait_event(async_done, lowest_in_progress(domain) >= cookie);
+	for (;;) {
+		prepare_to_wait(&async_done, &await.wait, TASK_UNINTERRUPTIBLE);
+
+		if (lowest_in_progress(domain) >= cookie)
+			break;
+		schedule();
+	}
+	finish_wait(&async_done, &await.wait);
 
 	pr_debug("async_continuing @ %i after %lli usec\n", task_pid_nr(current),
 		 microseconds_since(starttime));
-- 
2.39.3


  reply	other threads:[~2025-06-25 20:19 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-06-25 20:18 [PATCH v10 0/5] shut down devices asynchronously Stuart Hayes
2025-06-25 20:18 ` Stuart Hayes [this message]
2025-07-08 22:17   ` [PATCH v10 1/5] kernel/async: streamline cookie synchronization Sultan Alsawaf
2025-06-25 20:18 ` [PATCH v10 2/5] driver core: don't always lock parent in shutdown Stuart Hayes
2025-07-01  8:50   ` Greg Kroah-Hartman
2025-07-02 14:38     ` David Jeffery
2025-06-25 20:18 ` [PATCH v10 3/5] driver core: separate function to shutdown one device Stuart Hayes
2025-06-25 20:18 ` [PATCH v10 4/5] driver core: shut down devices asynchronously Stuart Hayes
2025-06-25 20:18 ` [PATCH v10 5/5] nvme-pci: Make driver prefer asynchronous shutdown Stuart Hayes
2025-06-30 20:33 ` [PATCH v10 0/5] shut down devices asynchronously Michael Kelley
2025-06-30 22:02   ` Laurence Oberman
2025-07-03 11:46 ` Christoph Hellwig
2025-07-03 15:41   ` Jeremy Allison
2025-07-04 13:45     ` David Jeffery
2025-07-04 16:26       ` Sultan Alsawaf
2025-07-07 15:34         ` David Jeffery
2025-07-07 20:49           ` stuart hayes
2025-07-08  0:00             ` Sultan Alsawaf
2025-07-08 21:47               ` Sultan Alsawaf
2025-07-08 21:31             ` Sultan Alsawaf
2025-07-03 15:59   ` stuart hayes
2025-07-04 13:38   ` David Jeffery
2025-07-04 13:44     ` Greg Kroah-Hartman
2025-07-04 14:09       ` David Jeffery
2025-07-04 14:13         ` Greg Kroah-Hartman

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=20250625201853.84062-2-stuart.w.hayes@gmail.com \
    --to=stuart.w.hayes@gmail.com \
    --cc=Martin.Belanger@dell.com \
    --cc=axboe@fb.com \
    --cc=djeffery@redhat.com \
    --cc=dwagner@suse.de \
    --cc=gregkh@linuxfoundation.org \
    --cc=hch@lst.de \
    --cc=jallison@ciq.com \
    --cc=jan.kiszka@seimens.com \
    --cc=kbusch@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-nvme@lists.infradead.org \
    --cc=lukas@wunner.de \
    --cc=nathan@kernel.org \
    --cc=oohall@gmail.com \
    --cc=rafael@kernel.org \
    --cc=sagi@grimberg.me \
    --cc=spasswolf@web.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

all inboxes | Powered by JetHome®