From: Alex Kogan <alex.kogan@oracle.com>
To: linux@armlinux.org.uk, peterz@infradead.org, mingo@redhat.com,
will.deacon@arm.com, arnd@arndb.de, longman@redhat.com,
linux-arch@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
linux-kernel@vger.kernel.org, tglx@linutronix.de, bp@alien8.de,
hpa@zytor.com, x86@kernel.org, guohanjun@huawei.com,
jglauber@marvell.com
Cc: steven.sistare@oracle.com, daniel.m.jordan@oracle.com,
alex.kogan@oracle.com, dave.dice@oracle.com
Subject: [PATCH v15 4/6] locking/qspinlock: Introduce starvation avoidance into CNA
Date: Fri, 14 May 2021 16:07:41 -0400 [thread overview]
Message-ID: <20210514200743.3026725-5-alex.kogan@oracle.com> (raw)
In-Reply-To: <20210514200743.3026725-1-alex.kogan@oracle.com>
Keep track of the time the thread at the head of the secondary queue
has been waiting, and force inter-node handoff once this time passes
a preset threshold. The default value for the threshold (1ms) can be
overridden with the new kernel boot command-line option
"qspinlock.numa_spinlock_threshold_ns".
Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
.../admin-guide/kernel-parameters.txt | 8 ++
kernel/locking/qspinlock_cna.h | 81 +++++++++++++++----
2 files changed, 73 insertions(+), 16 deletions(-)
diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 94d35507560c..e2bcf6f4e048 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -4183,6 +4183,14 @@
[KNL] Number of legacy pty's. Overwrites compiled-in
default number.
+ qspinlock.numa_spinlock_threshold_ns= [NUMA, PV_OPS]
+ Set the time threshold in nanoseconds for the
+ number of intra-node lock hand-offs before the
+ NUMA-aware spinlock is forced to be passed to
+ a thread on another NUMA node. Smaller values
+ result in a more fair, but less performant spinlock,
+ and vice versa. The default value is 1000000 (=1ms).
+
quiet [KNL] Disable most log messages
r128= [HW,DRM]
diff --git a/kernel/locking/qspinlock_cna.h b/kernel/locking/qspinlock_cna.h
index ca564e64e5de..0b991c340fb1 100644
--- a/kernel/locking/qspinlock_cna.h
+++ b/kernel/locking/qspinlock_cna.h
@@ -4,6 +4,8 @@
#endif
#include <linux/topology.h>
+#include <linux/sched/clock.h>
+#include <linux/moduleparam.h>
/*
* Implement a NUMA-aware version of MCS (aka CNA, or compact NUMA-aware lock).
@@ -37,19 +39,39 @@
* gradually filter the primary queue, leaving only waiters running on the same
* preferred NUMA node.
*
+ * We change the NUMA node preference after a waiter at the head of the
+ * secondary queue spins for a certain amount of time (1ms, by default).
+ * We do that by flushing the secondary queue into the head of the primary queue,
+ * effectively changing the preference to the NUMA node of the waiter at the head
+ * of the secondary queue at the time of the flush.
+ *
* For more details, see https://arxiv.org/abs/1810.05600.
*
* Authors: Alex Kogan <alex.kogan@oracle.com>
* Dave Dice <dave.dice@oracle.com>
*/
+#define FLUSH_SECONDARY_QUEUE 1
+
struct cna_node {
struct mcs_spinlock mcs;
u16 numa_node;
u16 real_numa_node;
u32 encoded_tail; /* self */
+ u64 start_time;
};
+static ulong numa_spinlock_threshold_ns = 1000000; /* 1ms, by default */
+module_param(numa_spinlock_threshold_ns, ulong, 0644);
+
+static inline bool intra_node_threshold_reached(struct cna_node *cn)
+{
+ u64 current_time = local_clock();
+ u64 threshold = cn->start_time + numa_spinlock_threshold_ns;
+
+ return current_time > threshold;
+}
+
static void __init cna_init_nodes_per_cpu(unsigned int cpu)
{
struct mcs_spinlock *base = per_cpu_ptr(&qnodes[0].mcs, cpu);
@@ -92,6 +114,7 @@ static __always_inline void cna_init_node(struct mcs_spinlock *node)
struct cna_node *cn = (struct cna_node *)node;
cn->numa_node = cn->real_numa_node;
+ cn->start_time = 0;
}
/*
@@ -191,8 +214,14 @@ static void cna_splice_next(struct mcs_spinlock *node,
/* stick `next` on the secondary queue tail */
if (node->locked <= 1) { /* if secondary queue is empty */
+ struct cna_node *cn = (struct cna_node *)node;
+
/* create secondary queue */
next->next = next;
+
+ cn->start_time = local_clock();
+ /* secondary queue is not empty iff start_time != 0 */
+ WARN_ON(!cn->start_time);
} else {
/* add to the tail of the secondary queue */
struct mcs_spinlock *tail_2nd = decode_tail(node->locked);
@@ -240,12 +269,18 @@ static int cna_order_queue(struct mcs_spinlock *node)
static __always_inline u32 cna_wait_head_or_lock(struct qspinlock *lock,
struct mcs_spinlock *node)
{
- /*
- * Try and put the time otherwise spent spin waiting on
- * _Q_LOCKED_PENDING_MASK to use by sorting our lists.
- */
- while (LOCK_IS_BUSY(lock) && !cna_order_queue(node))
- cpu_relax();
+ struct cna_node *cn = (struct cna_node *)node;
+
+ if (!cn->start_time || !intra_node_threshold_reached(cn)) {
+ /*
+ * Try and put the time otherwise spent spin waiting on
+ * _Q_LOCKED_PENDING_MASK to use by sorting our lists.
+ */
+ while (LOCK_IS_BUSY(lock) && !cna_order_queue(node))
+ cpu_relax();
+ } else {
+ cn->start_time = FLUSH_SECONDARY_QUEUE;
+ }
return 0; /* we lied; we didn't wait, go do so now */
}
@@ -253,24 +288,38 @@ static __always_inline u32 cna_wait_head_or_lock(struct qspinlock *lock,
static inline void cna_lock_handoff(struct mcs_spinlock *node,
struct mcs_spinlock *next)
{
+ struct cna_node *cn = (struct cna_node *)node;
u32 val = 1;
- if (node->locked > 1) {
- struct cna_node *cn = (struct cna_node *)node;
+ if (cn->start_time != FLUSH_SECONDARY_QUEUE) {
+ if (node->locked > 1) {
+ val = node->locked; /* preseve secondary queue */
+
+ /*
+ * We have a local waiter, either real or fake one;
+ * reload @next in case it was changed by cna_order_queue().
+ */
+ next = node->next;
- val = node->locked; /* preseve secondary queue */
+ /*
+ * Pass over NUMA node id of primary queue, to maintain the
+ * preference even if the next waiter is on a different node.
+ */
+ ((struct cna_node *)next)->numa_node = cn->numa_node;
+ ((struct cna_node *)next)->start_time = cn->start_time;
+ }
+ } else {
/*
- * We have a local waiter, either real or fake one;
- * reload @next in case it was changed by cna_order_queue().
+ * We decided to flush the secondary queue;
+ * this can only happen if that queue is not empty.
*/
- next = node->next;
-
+ WARN_ON(node->locked <= 1);
/*
- * Pass over NUMA node id of primary queue, to maintain the
- * preference even if the next waiter is on a different node.
+ * Splice the secondary queue onto the primary queue and pass the lock
+ * to the longest waiting remote waiter.
*/
- ((struct cna_node *)next)->numa_node = cn->numa_node;
+ next = cna_splice_head(NULL, 0, node, next);
}
arch_mcs_lock_handoff(&next->locked, val);
--
2.24.3 (Apple Git-128)
next prev parent reply other threads:[~2021-05-14 20:09 UTC|newest]
Thread overview: 28+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-14 20:07 [PATCH v15 0/6] Add NUMA-awareness to qspinlock Alex Kogan
2021-05-14 20:07 ` [PATCH v15 1/6] locking/qspinlock: Rename mcs lock/unlock macros and make them more generic Alex Kogan
2021-05-14 20:07 ` [PATCH v15 2/6] locking/qspinlock: Refactor the qspinlock slow path Alex Kogan
2021-05-14 20:07 ` [PATCH v15 3/6] locking/qspinlock: Introduce CNA into the slow path of qspinlock Alex Kogan
2021-09-22 19:25 ` Davidlohr Bueso
2021-09-22 19:52 ` Waiman Long
2023-08-04 1:49 ` Guo Ren
2021-09-30 10:05 ` Barry Song
2023-08-02 23:14 ` Guo Ren
2023-08-03 8:50 ` Peter Zijlstra
2023-08-03 10:28 ` Guo Ren
2023-08-03 11:56 ` Peter Zijlstra
2023-08-04 1:33 ` Guo Ren
2023-08-04 1:38 ` Guo Ren
2023-08-04 8:25 ` Peter Zijlstra
2023-08-04 14:17 ` Guo Ren
2023-08-04 18:23 ` Peter Zijlstra
2023-08-05 0:19 ` Guo Ren
2021-05-14 20:07 ` Alex Kogan [this message]
2021-05-14 20:07 ` [PATCH v15 5/6] locking/qspinlock: Avoid moving certain threads between waiting queues in CNA Alex Kogan
2021-05-14 20:07 ` [PATCH v15 6/6] locking/qspinlock: Introduce the shuffle reduction optimization into CNA Alex Kogan
2021-09-30 9:44 ` [PATCH v15 0/6] Add NUMA-awareness to qspinlock Barry Song
2021-09-30 16:58 ` Waiman Long
2021-09-30 22:57 ` Barry Song
2021-09-30 23:51 ` Alex Kogan
2021-12-13 20:37 ` Alex Kogan
2021-12-15 15:13 ` Alex Kogan
2022-04-11 17:09 ` Alex Kogan
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=20210514200743.3026725-5-alex.kogan@oracle.com \
--to=alex.kogan@oracle.com \
--cc=arnd@arndb.de \
--cc=bp@alien8.de \
--cc=daniel.m.jordan@oracle.com \
--cc=dave.dice@oracle.com \
--cc=guohanjun@huawei.com \
--cc=hpa@zytor.com \
--cc=jglauber@marvell.com \
--cc=linux-arch@vger.kernel.org \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux@armlinux.org.uk \
--cc=longman@redhat.com \
--cc=mingo@redhat.com \
--cc=peterz@infradead.org \
--cc=steven.sistare@oracle.com \
--cc=tglx@linutronix.de \
--cc=will.deacon@arm.com \
--cc=x86@kernel.org \
/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;
as well as URLs for NNTP newsgroup(s).