[PATCH v12 0/5] Add NUMA-awareness to qspinlock

[PATCH v12 0/5] Add NUMA-awareness to qspinlock

[Alex Kogan]

Minor change from v11:
----------------

Fix documentation issue, as requested by Randy Dunlap and Longman.
The rest of the series is unchanged.

Summary
-------

Lock throughput can be increased by handing a lock to a waiter on the
same NUMA node as the lock holder, provided care is taken to avoid
starvation of waiters on other NUMA nodes. This patch introduces CNA
(compact NUMA-aware lock) as the slow path for qspinlock. It is
enabled through a configuration option (NUMA_AWARE_SPINLOCKS).

CNA is a NUMA-aware version of the MCS lock. Spinning threads are
organized in two queues, a primary queue for threads running on the same
node as the current lock holder, and a secondary queue for threads
running on other nodes. Threads store the ID of the node on which
they are running in their queue nodes. After acquiring the MCS lock and
before acquiring the spinlock, the MCS lock holder checks whether the next
waiter in the primary queue (if exists) is running on the same NUMA node.
If it is not, that waiter is detached from the main queue and moved into
the tail of the secondary queue. This way, we gradually filter the primary
queue, leaving only waiters running on the same preferred NUMA node. Note
that certain priortized waiters (e.g., in irq and nmi contexts) are
excluded from being moved to the secondary queue. We change the NUMA node
preference after a waiter at the head of the secondary queue spins for a
certain amount of time. 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.

More details are available at https://arxiv.org/abs/1810.05600.

We have done some performance evaluation with the locktorture module
as well as with several benchmarks from the will-it-scale repo.
The following locktorture results are from an Oracle X5-4 server
(four Intel Xeon E7-8895 v3 @ 2.60GHz sockets with 18 hyperthreaded
cores each). Each number represents an average (over 25 runs) of the
total number of ops (x10^7) reported at the end of each run. The 
standard deviation is also reported in (), and in general is about 3%
from the average. The 'stock' kernel is v5.10.0-rc2,
commit 4ef8451b3326, compiled in the default configuration. 
'patch-CNA' is the modified kernel with NUMA_AWARE_SPINLOCKS set; 
the speedup is calculated dividing 'patch-CNA' by 'stock'.

#thr  	 stock        patch-CNA   speedup (patch-CNA/stock)
  1  2.672 (0.109)  2.679 (0.110)  1.002
  2  3.247 (0.161)  3.276 (0.136)  1.009
  4  4.376 (0.139)  4.820 (0.181)  1.101
  8  5.134 (0.137)  7.125 (0.164)  1.388
 16  5.875 (0.113)  8.903 (0.209)  1.515
 32  6.298 (0.105)  9.911 (0.254)  1.574
 36  6.439 (0.125)  9.972 (0.226)  1.549
 72  6.249 (0.109)  10.375 (0.209)  1.660
108  6.082 (0.063)  10.511 (0.190)  1.728
142  5.822 (0.058)  10.448 (0.177)  1.795

The following tables contain throughput results (ops/us) from the same
setup for will-it-scale/open1_threads: 

#thr  	 stock        patch-CNA   speedup (patch-CNA/stock)
  1  0.508 (0.002)  0.507 (0.002)  0.999
  2  0.759 (0.016)  0.767 (0.019)  1.011
  4  1.397 (0.032)  1.411 (0.029)  1.011
  8  1.694 (0.079)  1.656 (0.120)  0.978
 16  1.867 (0.107)  1.809 (0.121)  0.969
 32  1.006 (0.056)  1.752 (0.093)  1.742
 36  0.934 (0.099)  1.724 (0.064)  1.846
 72  0.804 (0.045)  1.632 (0.073)  2.030
108  0.828 (0.036)  1.690 (0.065)  2.041
142  0.784 (0.035)  1.701 (0.074)  2.168

and will-it-scale/lock2_threads:

#thr  	 stock        patch-CNA   speedup (patch-CNA/stock)
  1  1.590 (0.004)  1.603 (0.005)  1.008
  2  2.802 (0.057)  2.802 (0.063)  1.000
  4  5.478 (0.144)  5.396 (0.299)  0.985
  8  4.166 (0.304)  4.131 (0.402)  0.992
 16  4.147 (0.137)  3.983 (0.173)  0.961
 32  2.492 (0.067)  3.888 (0.125)  1.560
 36  2.471 (0.094)  3.908 (0.112)  1.581
 72  1.886 (0.092)  3.926 (0.106)  2.081
108  1.883 (0.101)  3.935 (0.096)  2.089
142  1.801 (0.112)  3.907 (0.111)  2.169

Our evaluation shows that CNA also improves performance of user 
applications that have hot pthread mutexes. Those mutexes are 
blocking, and waiting threads park and unpark via the futex 
mechanism in the kernel. Given that kernel futex chains, which
are hashed by the mutex address, are each protected by a 
chain-specific spin lock, the contention on a user-mode mutex 
translates into contention on a kernel level spinlock. 

Here are the throughput results (ops/us) for the leveldb ‘readrandom’
benchmark:

#thr  	 stock        patch-CNA   speedup (patch-CNA/stock)
  1  0.532 (0.013)  0.533 (0.023)  1.004
  2  0.836 (0.050)  0.843 (0.031)  1.009
  4  1.039 (0.163)  1.087 (0.151)  1.046
  8  1.095 (0.181)  1.178 (0.165)  1.076
 16  1.002 (0.144)  1.196 (0.019)  1.194
 32  0.726 (0.034)  1.163 (0.026)  1.601
 36  0.691 (0.030)  1.163 (0.020)  1.683
 72  0.627 (0.014)  1.136 (0.022)  1.812
108  0.613 (0.014)  1.143 (0.023)  1.865
142  0.610 (0.014)  1.120 (0.018)  1.838

Further comments are welcome and appreciated.

Alex Kogan (5):
  locking/qspinlock: Rename mcs lock/unlock macros and make them more
    generic
  locking/qspinlock: Refactor the qspinlock slow path
  locking/qspinlock: Introduce CNA into the slow path of qspinlock
  locking/qspinlock: Introduce starvation avoidance into CNA
  locking/qspinlock: Avoid moving certain threads between waiting queues
    in CNA

 Documentation/admin-guide/kernel-parameters.txt |  19 ++
 arch/arm/include/asm/mcs_spinlock.h             |   6 +-
 arch/x86/Kconfig                                |  20 ++
 arch/x86/include/asm/qspinlock.h                |   4 +
 arch/x86/kernel/alternative.c                   |   4 +
 include/asm-generic/mcs_spinlock.h              |   4 +-
 kernel/locking/mcs_spinlock.h                   |  20 +-
 kernel/locking/qspinlock.c                      |  82 ++++-
 kernel/locking/qspinlock_cna.h                  | 421 ++++++++++++++++++++++++
 kernel/locking/qspinlock_paravirt.h             |   2 +-
 10 files changed, 559 insertions(+), 23 deletions(-)
 create mode 100644 kernel/locking/qspinlock_cna.h

-- 
2.7.4

[PATCH v12 1/5] locking/qspinlock: Rename mcs lock/unlock macros and make them more generic

[Alex Kogan]

The mcs unlock macro (arch_mcs_lock_handoff) should accept the value to be
stored into the lock argument as another argument. This allows using the
same macro in cases where the value to be stored when passing the lock is
different from 1.

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 arch/arm/include/asm/mcs_spinlock.h |  6 +++---
 include/asm-generic/mcs_spinlock.h  |  4 ++--
 kernel/locking/mcs_spinlock.h       | 18 +++++++++---------
 kernel/locking/qspinlock.c          |  4 ++--
 kernel/locking/qspinlock_paravirt.h |  2 +-
 5 files changed, 17 insertions(+), 17 deletions(-)

diff --git a/arch/arm/include/asm/mcs_spinlock.h b/arch/arm/include/asm/mcs_spinlock.h
index 529d2cf..1eb4d73 100644
--- a/arch/arm/include/asm/mcs_spinlock.h
+++ b/arch/arm/include/asm/mcs_spinlock.h
@@ -6,7 +6,7 @@
 #include <asm/spinlock.h>
 
 /* MCS spin-locking. */
-#define arch_mcs_spin_lock_contended(lock)				\
+#define arch_mcs_spin_wait(lock)					\
 do {									\
 	/* Ensure prior stores are observed before we enter wfe. */	\
 	smp_mb();							\
@@ -14,9 +14,9 @@ do {									\
 		wfe();							\
 } while (0)								\
 
-#define arch_mcs_spin_unlock_contended(lock)				\
+#define arch_mcs_lock_handoff(lock, val)				\
 do {									\
-	smp_store_release(lock, 1);					\
+	smp_store_release((lock), (val));				\
 	dsb_sev();							\
 } while (0)
 
diff --git a/include/asm-generic/mcs_spinlock.h b/include/asm-generic/mcs_spinlock.h
index 10cd4ff..f933d99 100644
--- a/include/asm-generic/mcs_spinlock.h
+++ b/include/asm-generic/mcs_spinlock.h
@@ -4,8 +4,8 @@
 /*
  * Architectures can define their own:
  *
- *   arch_mcs_spin_lock_contended(l)
- *   arch_mcs_spin_unlock_contended(l)
+ *   arch_mcs_spin_wait(l)
+ *   arch_mcs_lock_handoff(l, val)
  *
  * See kernel/locking/mcs_spinlock.c.
  */
diff --git a/kernel/locking/mcs_spinlock.h b/kernel/locking/mcs_spinlock.h
index 5e10153..904ba5d 100644
--- a/kernel/locking/mcs_spinlock.h
+++ b/kernel/locking/mcs_spinlock.h
@@ -21,7 +21,7 @@ struct mcs_spinlock {
 	int count;  /* nesting count, see qspinlock.c */
 };
 
-#ifndef arch_mcs_spin_lock_contended
+#ifndef arch_mcs_spin_wait
 /*
  * Using smp_cond_load_acquire() provides the acquire semantics
  * required so that subsequent operations happen after the
@@ -29,20 +29,20 @@ struct mcs_spinlock {
  * ARM64 would like to do spin-waiting instead of purely
  * spinning, and smp_cond_load_acquire() provides that behavior.
  */
-#define arch_mcs_spin_lock_contended(l)					\
-do {									\
-	smp_cond_load_acquire(l, VAL);					\
+#define arch_mcs_spin_wait(l)					\
+do {								\
+	smp_cond_load_acquire(l, VAL);				\
 } while (0)
 #endif
 
-#ifndef arch_mcs_spin_unlock_contended
+#ifndef arch_mcs_lock_handoff
 /*
  * smp_store_release() provides a memory barrier to ensure all
  * operations in the critical section has been completed before
  * unlocking.
  */
-#define arch_mcs_spin_unlock_contended(l)				\
-	smp_store_release((l), 1)
+#define arch_mcs_lock_handoff(l, val)				\
+	smp_store_release((l), (val))
 #endif
 
 /*
@@ -91,7 +91,7 @@ void mcs_spin_lock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
 	WRITE_ONCE(prev->next, node);
 
 	/* Wait until the lock holder passes the lock down. */
-	arch_mcs_spin_lock_contended(&node->locked);
+	arch_mcs_spin_wait(&node->locked);
 }
 
 /*
@@ -115,7 +115,7 @@ void mcs_spin_unlock(struct mcs_spinlock **lock, struct mcs_spinlock *node)
 	}
 
 	/* Pass lock to next waiter. */
-	arch_mcs_spin_unlock_contended(&next->locked);
+	arch_mcs_lock_handoff(&next->locked, 1);
 }
 
 #endif /* __LINUX_MCS_SPINLOCK_H */
diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index cbff6ba..435d696 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -471,7 +471,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 		WRITE_ONCE(prev->next, node);
 
 		pv_wait_node(node, prev);
-		arch_mcs_spin_lock_contended(&node->locked);
+		arch_mcs_spin_wait(&node->locked);
 
 		/*
 		 * While waiting for the MCS lock, the next pointer may have
@@ -550,7 +550,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 	if (!next)
 		next = smp_cond_load_relaxed(&node->next, (VAL));
 
-	arch_mcs_spin_unlock_contended(&next->locked);
+	arch_mcs_lock_handoff(&next->locked, 1);
 	pv_kick_node(lock, next);
 
 release:
diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
index e84d21a..619d80f 100644
--- a/kernel/locking/qspinlock_paravirt.h
+++ b/kernel/locking/qspinlock_paravirt.h
@@ -368,7 +368,7 @@ static void pv_kick_node(struct qspinlock *lock, struct mcs_spinlock *node)
 	 *
 	 * Matches with smp_store_mb() and cmpxchg() in pv_wait_node()
 	 *
-	 * The write to next->locked in arch_mcs_spin_unlock_contended()
+	 * The write to next->locked in arch_mcs_lock_handoff()
 	 * must be ordered before the read of pn->state in the cmpxchg()
 	 * below for the code to work correctly. To guarantee full ordering
 	 * irrespective of the success or failure of the cmpxchg(),
-- 
2.7.4

[PATCH v12 2/5] locking/qspinlock: Refactor the qspinlock slow path

[Alex Kogan]

Move some of the code manipulating the spin lock into separate functions.
This would allow easier integration of alternative ways to manipulate
that lock.

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 kernel/locking/qspinlock.c | 38 ++++++++++++++++++++++++++++++++++++--
 1 file changed, 36 insertions(+), 2 deletions(-)

diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index 435d696..e351870 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -289,6 +289,34 @@ static __always_inline u32  __pv_wait_head_or_lock(struct qspinlock *lock,
 #define queued_spin_lock_slowpath	native_queued_spin_lock_slowpath
 #endif
 
+/*
+ * __try_clear_tail - try to clear tail by setting the lock value to
+ * _Q_LOCKED_VAL.
+ * @lock: Pointer to the queued spinlock structure
+ * @val: Current value of the lock
+ * @node: Pointer to the MCS node of the lock holder
+ */
+static __always_inline bool __try_clear_tail(struct qspinlock *lock,
+					     u32 val,
+					     struct mcs_spinlock *node)
+{
+	return atomic_try_cmpxchg_relaxed(&lock->val, &val, _Q_LOCKED_VAL);
+}
+
+/*
+ * __mcs_lock_handoff - pass the MCS lock to the next waiter
+ * @node: Pointer to the MCS node of the lock holder
+ * @next: Pointer to the MCS node of the first waiter in the MCS queue
+ */
+static __always_inline void __mcs_lock_handoff(struct mcs_spinlock *node,
+					       struct mcs_spinlock *next)
+{
+	arch_mcs_lock_handoff(&next->locked, 1);
+}
+
+#define try_clear_tail		__try_clear_tail
+#define mcs_lock_handoff	__mcs_lock_handoff
+
 #endif /* _GEN_PV_LOCK_SLOWPATH */
 
 /**
@@ -533,7 +561,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 	 *       PENDING will make the uncontended transition fail.
 	 */
 	if ((val & _Q_TAIL_MASK) == tail) {
-		if (atomic_try_cmpxchg_relaxed(&lock->val, &val, _Q_LOCKED_VAL))
+		if (try_clear_tail(lock, val, node))
 			goto release; /* No contention */
 	}
 
@@ -550,7 +578,7 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 	if (!next)
 		next = smp_cond_load_relaxed(&node->next, (VAL));
 
-	arch_mcs_lock_handoff(&next->locked, 1);
+	mcs_lock_handoff(node, next);
 	pv_kick_node(lock, next);
 
 release:
@@ -575,6 +603,12 @@ EXPORT_SYMBOL(queued_spin_lock_slowpath);
 #undef pv_kick_node
 #undef pv_wait_head_or_lock
 
+#undef try_clear_tail
+#define try_clear_tail		__try_clear_tail
+
+#undef mcs_lock_handoff
+#define mcs_lock_handoff			__mcs_lock_handoff
+
 #undef  queued_spin_lock_slowpath
 #define queued_spin_lock_slowpath	__pv_queued_spin_lock_slowpath
 
-- 
2.7.4

[PATCH v12 3/5] locking/qspinlock: Introduce CNA into the slow path of qspinlock

[Alex Kogan]

In CNA, spinning threads are organized in two queues, a primary queue for
threads running on the same node as the current lock holder, and a
secondary queue for threads running on other nodes. After acquiring the
MCS lock and before acquiring the spinlock, the MCS lock
holder checks whether the next waiter in the primary queue (if exists) is
running on the same NUMA node. If it is not, that waiter is detached from
the main queue and moved into the tail of the secondary queue. This way,
we gradually filter the primary queue, leaving only waiters running on
the same preferred NUMA node. For more details, see
https://arxiv.org/abs/1810.05600.

Note that this variant of CNA may introduce starvation by continuously
passing the lock between waiters in the main queue. This issue will be
addressed later in the series.

Enabling CNA is controlled via a new configuration option
(NUMA_AWARE_SPINLOCKS). By default, the CNA variant is patched in at the
boot time only if we run on a multi-node machine in native environment and
the new config is enabled. (For the time being, the patching requires
CONFIG_PARAVIRT_SPINLOCKS to be enabled as well. However, this should be
resolved once static_call() is available.) This default behavior can be
overridden with the new kernel boot command-line option
"numa_spinlock=on/off" (default is "auto").

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  10 +
 arch/x86/Kconfig                                |  20 ++
 arch/x86/include/asm/qspinlock.h                |   4 +
 arch/x86/kernel/alternative.c                   |   4 +
 kernel/locking/mcs_spinlock.h                   |   2 +-
 kernel/locking/qspinlock.c                      |  42 ++-
 kernel/locking/qspinlock_cna.h                  | 336 ++++++++++++++++++++++++
 7 files changed, 413 insertions(+), 5 deletions(-)
 create mode 100644 kernel/locking/qspinlock_cna.h

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 526d65d..41ee8ff 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -3423,6 +3423,16 @@
 	numa_balancing=	[KNL,X86] Enable or disable automatic NUMA balancing.
 			Allowed values are enable and disable
 
+	numa_spinlock=	[NUMA, PV_OPS] Select the NUMA-aware variant
+			of spinlock. The options are:
+			auto - Enable this variant if running on a multi-node
+			machine in native environment.
+			on  - Unconditionally enable this variant.
+			off - Unconditionally disable this variant.
+
+			Not specifying this option is equivalent to
+			numa_spinlock=auto.
+
 	numa_zonelist_order= [KNL, BOOT] Select zonelist order for NUMA.
 			'node', 'default' can be specified
 			This can be set from sysctl after boot.
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index f6946b8..6a04cf61 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1564,6 +1564,26 @@ config NUMA
 
 	  Otherwise, you should say N.
 
+config NUMA_AWARE_SPINLOCKS
+	bool "Numa-aware spinlocks"
+	depends on NUMA
+	depends on QUEUED_SPINLOCKS
+	depends on 64BIT
+	# For now, we depend on PARAVIRT_SPINLOCKS to make the patching work.
+	# This is awkward, but hopefully would be resolved once static_call()
+	# is available.
+	depends on PARAVIRT_SPINLOCKS
+	default y
+	help
+	  Introduce NUMA (Non Uniform Memory Access) awareness into
+	  the slow path of spinlocks.
+
+	  In this variant of qspinlock, the kernel will try to keep the lock
+	  on the same node, thus reducing the number of remote cache misses,
+	  while trading some of the short term fairness for better performance.
+
+	  Say N if you want absolute first come first serve fairness.
+
 config AMD_NUMA
 	def_bool y
 	prompt "Old style AMD Opteron NUMA detection"
diff --git a/arch/x86/include/asm/qspinlock.h b/arch/x86/include/asm/qspinlock.h
index d86ab94..21d09e8 100644
--- a/arch/x86/include/asm/qspinlock.h
+++ b/arch/x86/include/asm/qspinlock.h
@@ -27,6 +27,10 @@ static __always_inline u32 queued_fetch_set_pending_acquire(struct qspinlock *lo
 	return val;
 }
 
+#ifdef CONFIG_NUMA_AWARE_SPINLOCKS
+extern void cna_configure_spin_lock_slowpath(void);
+#endif
+
 #ifdef CONFIG_PARAVIRT_SPINLOCKS
 extern void native_queued_spin_lock_slowpath(struct qspinlock *lock, u32 val);
 extern void __pv_init_lock_hash(void);
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index 2400ad6..e04f48c 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -741,6 +741,10 @@ void __init alternative_instructions(void)
 	}
 #endif
 
+#if defined(CONFIG_NUMA_AWARE_SPINLOCKS)
+	cna_configure_spin_lock_slowpath();
+#endif
+
 	apply_paravirt(__parainstructions, __parainstructions_end);
 
 	restart_nmi();
diff --git a/kernel/locking/mcs_spinlock.h b/kernel/locking/mcs_spinlock.h
index 904ba5d..5e47ffb 100644
--- a/kernel/locking/mcs_spinlock.h
+++ b/kernel/locking/mcs_spinlock.h
@@ -17,7 +17,7 @@
 
 struct mcs_spinlock {
 	struct mcs_spinlock *next;
-	int locked; /* 1 if lock acquired */
+	unsigned int locked; /* 1 if lock acquired */
 	int count;  /* nesting count, see qspinlock.c */
 };
 
diff --git a/kernel/locking/qspinlock.c b/kernel/locking/qspinlock.c
index e351870..8c1a21b 100644
--- a/kernel/locking/qspinlock.c
+++ b/kernel/locking/qspinlock.c
@@ -11,7 +11,7 @@
  *          Peter Zijlstra <peterz@infradead.org>
  */
 
-#ifndef _GEN_PV_LOCK_SLOWPATH
+#if !defined(_GEN_PV_LOCK_SLOWPATH) && !defined(_GEN_CNA_LOCK_SLOWPATH)
 
 #include <linux/smp.h>
 #include <linux/bug.h>
@@ -71,7 +71,8 @@
 /*
  * On 64-bit architectures, the mcs_spinlock structure will be 16 bytes in
  * size and four of them will fit nicely in one 64-byte cacheline. For
- * pvqspinlock, however, we need more space for extra data. To accommodate
+ * pvqspinlock, however, we need more space for extra data. The same also
+ * applies for the NUMA-aware variant of spinlocks (CNA). To accommodate
  * that, we insert two more long words to pad it up to 32 bytes. IOW, only
  * two of them can fit in a cacheline in this case. That is OK as it is rare
  * to have more than 2 levels of slowpath nesting in actual use. We don't
@@ -80,7 +81,7 @@
  */
 struct qnode {
 	struct mcs_spinlock mcs;
-#ifdef CONFIG_PARAVIRT_SPINLOCKS
+#if defined(CONFIG_PARAVIRT_SPINLOCKS) || defined(CONFIG_NUMA_AWARE_SPINLOCKS)
 	long reserved[2];
 #endif
 };
@@ -104,6 +105,8 @@ struct qnode {
  * Exactly fits one 64-byte cacheline on a 64-bit architecture.
  *
  * PV doubles the storage and uses the second cacheline for PV state.
+ * CNA also doubles the storage and uses the second cacheline for
+ * CNA-specific state.
  */
 static DEFINE_PER_CPU_ALIGNED(struct qnode, qnodes[MAX_NODES]);
 
@@ -317,7 +320,7 @@ static __always_inline void __mcs_lock_handoff(struct mcs_spinlock *node,
 #define try_clear_tail		__try_clear_tail
 #define mcs_lock_handoff	__mcs_lock_handoff
 
-#endif /* _GEN_PV_LOCK_SLOWPATH */
+#endif /* _GEN_PV_LOCK_SLOWPATH && _GEN_CNA_LOCK_SLOWPATH */
 
 /**
  * queued_spin_lock_slowpath - acquire the queued spinlock
@@ -590,6 +593,37 @@ void queued_spin_lock_slowpath(struct qspinlock *lock, u32 val)
 EXPORT_SYMBOL(queued_spin_lock_slowpath);
 
 /*
+ * Generate the code for NUMA-aware spinlocks
+ */
+#if !defined(_GEN_CNA_LOCK_SLOWPATH) && defined(CONFIG_NUMA_AWARE_SPINLOCKS)
+#define _GEN_CNA_LOCK_SLOWPATH
+
+#undef pv_init_node
+#define pv_init_node			cna_init_node
+
+#undef pv_wait_head_or_lock
+#define pv_wait_head_or_lock		cna_wait_head_or_lock
+
+#undef try_clear_tail
+#define try_clear_tail			cna_try_clear_tail
+
+#undef mcs_lock_handoff
+#define mcs_lock_handoff			cna_lock_handoff
+
+#undef  queued_spin_lock_slowpath
+/*
+ * defer defining queued_spin_lock_slowpath until after the include to
+ * avoid a name clash with the identically named field in pv_ops.lock
+ * (see cna_configure_spin_lock_slowpath())
+ */
+#include "qspinlock_cna.h"
+#define queued_spin_lock_slowpath	__cna_queued_spin_lock_slowpath
+
+#include "qspinlock.c"
+
+#endif
+
+/*
  * Generate the paravirt code for queued_spin_unlock_slowpath().
  */
 #if !defined(_GEN_PV_LOCK_SLOWPATH) && defined(CONFIG_PARAVIRT_SPINLOCKS)
diff --git a/kernel/locking/qspinlock_cna.h b/kernel/locking/qspinlock_cna.h
new file mode 100644
index 0000000..590402a
--- /dev/null
+++ b/kernel/locking/qspinlock_cna.h
@@ -0,0 +1,336 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _GEN_CNA_LOCK_SLOWPATH
+#error "do not include this file"
+#endif
+
+#include <linux/topology.h>
+
+/*
+ * Implement a NUMA-aware version of MCS (aka CNA, or compact NUMA-aware lock).
+ *
+ * In CNA, spinning threads are organized in two queues, a primary queue for
+ * threads running on the same NUMA node as the current lock holder, and a
+ * secondary queue for threads running on other nodes. Schematically, it
+ * looks like this:
+ *
+ *    cna_node
+ *   +----------+     +--------+         +--------+
+ *   |mcs:next  | --> |mcs:next| --> ... |mcs:next| --> NULL  [Primary queue]
+ *   |mcs:locked| -.  +--------+         +--------+
+ *   +----------+  |
+ *                 `----------------------.
+ *                                        v
+ *                 +--------+         +--------+
+ *                 |mcs:next| --> ... |mcs:next|            [Secondary queue]
+ *                 +--------+         +--------+
+ *                     ^                    |
+ *                     `--------------------'
+ *
+ * N.B. locked := 1 if secondary queue is absent. Otherwise, it contains the
+ * encoded pointer to the tail of the secondary queue, which is organized as a
+ * circular list.
+ *
+ * After acquiring the MCS lock and before acquiring the spinlock, the MCS lock
+ * holder checks whether the next waiter in the primary queue (if exists) is
+ * running on the same NUMA node. If it is not, that waiter is detached from the
+ * main queue and moved into the tail of the secondary queue. This way, we
+ * gradually filter the primary queue, leaving only waiters running on the same
+ * preferred NUMA node.
+ *
+ * For more details, see https://arxiv.org/abs/1810.05600.
+ *
+ * Authors: Alex Kogan <alex.kogan@oracle.com>
+ *          Dave Dice <dave.dice@oracle.com>
+ */
+
+struct cna_node {
+	struct mcs_spinlock	mcs;
+	u16			numa_node;
+	u16			real_numa_node;
+	u32			encoded_tail;	/* self */
+	u32			partial_order;	/* enum val */
+};
+
+enum {
+	LOCAL_WAITER_FOUND,
+	LOCAL_WAITER_NOT_FOUND,
+};
+
+static void __init cna_init_nodes_per_cpu(unsigned int cpu)
+{
+	struct mcs_spinlock *base = per_cpu_ptr(&qnodes[0].mcs, cpu);
+	int numa_node = cpu_to_node(cpu);
+	int i;
+
+	for (i = 0; i < MAX_NODES; i++) {
+		struct cna_node *cn = (struct cna_node *)grab_mcs_node(base, i);
+
+		cn->real_numa_node = numa_node;
+		cn->encoded_tail = encode_tail(cpu, i);
+		/*
+		 * make sure @encoded_tail is not confused with other valid
+		 * values for @locked (0 or 1)
+		 */
+		WARN_ON(cn->encoded_tail <= 1);
+	}
+}
+
+static int __init cna_init_nodes(void)
+{
+	unsigned int cpu;
+
+	/*
+	 * this will break on 32bit architectures, so we restrict
+	 * the use of CNA to 64bit only (see arch/x86/Kconfig)
+	 */
+	BUILD_BUG_ON(sizeof(struct cna_node) > sizeof(struct qnode));
+	/* we store an ecoded tail word in the node's @locked field */
+	BUILD_BUG_ON(sizeof(u32) > sizeof(unsigned int));
+
+	for_each_possible_cpu(cpu)
+		cna_init_nodes_per_cpu(cpu);
+
+	return 0;
+}
+
+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;
+}
+
+/*
+ * cna_splice_head -- splice the entire secondary queue onto the head of the
+ * primary queue.
+ *
+ * Returns the new primary head node or NULL on failure.
+ */
+static struct mcs_spinlock *
+cna_splice_head(struct qspinlock *lock, u32 val,
+		struct mcs_spinlock *node, struct mcs_spinlock *next)
+{
+	struct mcs_spinlock *head_2nd, *tail_2nd;
+	u32 new;
+
+	tail_2nd = decode_tail(node->locked);
+	head_2nd = tail_2nd->next;
+
+	if (next) {
+		/*
+		 * If the primary queue is not empty, the primary tail doesn't
+		 * need to change and we can simply link the secondary tail to
+		 * the old primary head.
+		 */
+		tail_2nd->next = next;
+	} else {
+		/*
+		 * When the primary queue is empty, the secondary tail becomes
+		 * the primary tail.
+		 */
+
+		/*
+		 * Speculatively break the secondary queue's circular link such
+		 * that when the secondary tail becomes the primary tail it all
+		 * works out.
+		 */
+		tail_2nd->next = NULL;
+
+		/*
+		 * tail_2nd->next = NULL;	old = xchg_tail(lock, tail);
+		 *				prev = decode_tail(old);
+		 * try_cmpxchg_release(...);	WRITE_ONCE(prev->next, node);
+		 *
+		 * If the following cmpxchg() succeeds, our stores will not
+		 * collide.
+		 */
+		new = ((struct cna_node *)tail_2nd)->encoded_tail |
+			_Q_LOCKED_VAL;
+		if (!atomic_try_cmpxchg_release(&lock->val, &val, new)) {
+			/* Restore the secondary queue's circular link. */
+			tail_2nd->next = head_2nd;
+			return NULL;
+		}
+	}
+
+	/* The primary queue head now is what was the secondary queue head. */
+	return head_2nd;
+}
+
+static inline bool cna_try_clear_tail(struct qspinlock *lock, u32 val,
+				      struct mcs_spinlock *node)
+{
+	/*
+	 * We're here because the primary queue is empty; check the secondary
+	 * queue for remote waiters.
+	 */
+	if (node->locked > 1) {
+		struct mcs_spinlock *next;
+
+		/*
+		 * When there are waiters on the secondary queue, try to move
+		 * them back onto the primary queue and let them rip.
+		 */
+		next = cna_splice_head(lock, val, node, NULL);
+		if (next) {
+			arch_mcs_lock_handoff(&next->locked, 1);
+			return true;
+		}
+
+		return false;
+	}
+
+	/* Both queues are empty. Do what MCS does. */
+	return __try_clear_tail(lock, val, node);
+}
+
+/*
+ * cna_splice_tail -- splice the next node from the primary queue onto
+ * the secondary queue.
+ */
+static void cna_splice_next(struct mcs_spinlock *node,
+			    struct mcs_spinlock *next,
+			    struct mcs_spinlock *nnext)
+{
+	/* remove 'next' from the main queue */
+	node->next = nnext;
+
+	/* stick `next` on the secondary queue tail */
+	if (node->locked <= 1) { /* if secondary queue is empty */
+		/* create secondary queue */
+		next->next = next;
+	} else {
+		/* add to the tail of the secondary queue */
+		struct mcs_spinlock *tail_2nd = decode_tail(node->locked);
+		struct mcs_spinlock *head_2nd = tail_2nd->next;
+
+		tail_2nd->next = next;
+		next->next = head_2nd;
+	}
+
+	node->locked = ((struct cna_node *)next)->encoded_tail;
+}
+
+/*
+ * cna_order_queue - check whether the next waiter in the main queue is on
+ * the same NUMA node as the lock holder; if not, and it has a waiter behind
+ * it in the main queue, move the former onto the secondary queue.
+ */
+static u32 cna_order_queue(struct mcs_spinlock *node)
+{
+	struct mcs_spinlock *next = READ_ONCE(node->next);
+	int numa_node, next_numa_node;
+
+	if (!next)
+		return LOCAL_WAITER_NOT_FOUND;
+
+	numa_node = ((struct cna_node *)node)->numa_node;
+	next_numa_node = ((struct cna_node *)next)->numa_node;
+
+	if (next_numa_node != numa_node) {
+		struct mcs_spinlock *nnext = READ_ONCE(next->next);
+
+		if (nnext) {
+			cna_splice_next(node, next, nnext);
+			next = nnext;
+		}
+		/*
+		 * Inherit 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 = numa_node;
+	}
+	return LOCAL_WAITER_FOUND;
+}
+
+/* Abuse the pv_wait_head_or_lock() hook to get some work done */
+static __always_inline u32 cna_wait_head_or_lock(struct qspinlock *lock,
+						 struct mcs_spinlock *node)
+{
+	struct cna_node *cn = (struct cna_node *)node;
+
+	/*
+	 * Try and put the time otherwise spent spin waiting on
+	 * _Q_LOCKED_PENDING_MASK to use by sorting our lists.
+	 */
+	cn->partial_order = cna_order_queue(node);
+
+	return 0; /* we lied; we didn't wait, go do so now */
+}
+
+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;
+
+	u32 partial_order = cn->partial_order;
+
+	if (partial_order == LOCAL_WAITER_NOT_FOUND)
+		partial_order = cna_order_queue(node);
+
+	/*
+	 * At this point, we must have a successor in the main queue, so if we call
+	 * cna_order_queue() above, we will find a local waiter, either real or
+	 * fake one.
+	 */
+	WARN_ON(partial_order == LOCAL_WAITER_NOT_FOUND);
+
+	/*
+	 * We found a local waiter; reload @next in case it was changed by
+	 * cna_order_queue().
+	 */
+	next = node->next;
+	if (node->locked > 1)
+		val = node->locked;	/* preseve secondary queue */
+
+	arch_mcs_lock_handoff(&next->locked, val);
+}
+
+/*
+ * Constant (boot-param configurable) flag selecting the NUMA-aware variant
+ * of spinlock.  Possible values: -1 (off) / 0 (auto, default) / 1 (on).
+ */
+static int numa_spinlock_flag;
+
+static int __init numa_spinlock_setup(char *str)
+{
+	if (!strcmp(str, "auto")) {
+		numa_spinlock_flag = 0;
+		return 1;
+	} else if (!strcmp(str, "on")) {
+		numa_spinlock_flag = 1;
+		return 1;
+	} else if (!strcmp(str, "off")) {
+		numa_spinlock_flag = -1;
+		return 1;
+	}
+
+	return 0;
+}
+__setup("numa_spinlock=", numa_spinlock_setup);
+
+void __cna_queued_spin_lock_slowpath(struct qspinlock *lock, u32 val);
+
+/*
+ * Switch to the NUMA-friendly slow path for spinlocks when we have
+ * multiple NUMA nodes in native environment, unless the user has
+ * overridden this default behavior by setting the numa_spinlock flag.
+ */
+void __init cna_configure_spin_lock_slowpath(void)
+{
+
+	if (numa_spinlock_flag < 0)
+		return;
+
+	if (numa_spinlock_flag == 0 && (nr_node_ids < 2 ||
+		    pv_ops.lock.queued_spin_lock_slowpath !=
+			native_queued_spin_lock_slowpath))
+		return;
+
+	cna_init_nodes();
+
+	pv_ops.lock.queued_spin_lock_slowpath = __cna_queued_spin_lock_slowpath;
+
+	pr_info("Enabling CNA spinlock\n");
+}
-- 
2.7.4

[PATCH v12 4/5] locking/qspinlock: Introduce starvation avoidance into CNA

[Alex Kogan]

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 (10ms) can be
overridden with the new kernel boot command-line option
"numa_spinlock_threshold". The ms value is translated internally to the
nearest rounded-up jiffies.

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 Documentation/admin-guide/kernel-parameters.txt |  9 +++
 kernel/locking/qspinlock_cna.h                  | 95 +++++++++++++++++++++----
 2 files changed, 92 insertions(+), 12 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 41ee8ff..3bcf756 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -3433,6 +3433,15 @@
 			Not specifying this option is equivalent to
 			numa_spinlock=auto.
 
+	numa_spinlock_threshold=	[NUMA, PV_OPS]
+			Set the time threshold in milliseconds 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.	Valid values
+			are in the [1..100] range. Smaller values result
+			in a more fair, but less performant spinlock,
+			and vice versa. The default value is 10.
+
 	numa_zonelist_order= [KNL, BOOT] Select zonelist order for NUMA.
 			'node', 'default' can be specified
 			This can be set from sysctl after boot.
diff --git a/kernel/locking/qspinlock_cna.h b/kernel/locking/qspinlock_cna.h
index 590402a..d3e2754 100644
--- a/kernel/locking/qspinlock_cna.h
+++ b/kernel/locking/qspinlock_cna.h
@@ -37,6 +37,12 @@
  * 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 (10ms, 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>
@@ -49,13 +55,33 @@ struct cna_node {
 	u16			real_numa_node;
 	u32			encoded_tail;	/* self */
 	u32			partial_order;	/* enum val */
+	s32			start_time;
 };
 
 enum {
 	LOCAL_WAITER_FOUND,
 	LOCAL_WAITER_NOT_FOUND,
+	FLUSH_SECONDARY_QUEUE
 };
 
+/*
+ * Controls the threshold time in ms (default = 10) for intra-node lock
+ * hand-offs before the NUMA-aware variant of spinlock is forced to be
+ * passed to a thread on another NUMA node. The default setting can be
+ * changed with the "numa_spinlock_threshold" boot option.
+ */
+#define MSECS_TO_JIFFIES(m)	\
+	(((m) + (MSEC_PER_SEC / HZ) - 1) / (MSEC_PER_SEC / HZ))
+static int intra_node_handoff_threshold __ro_after_init = MSECS_TO_JIFFIES(10);
+
+static inline bool intra_node_threshold_reached(struct cna_node *cn)
+{
+	s32 current_time = (s32)jiffies;
+	s32 threshold = cn->start_time + intra_node_handoff_threshold;
+
+	return current_time - threshold > 0;
+}
+
 static void __init cna_init_nodes_per_cpu(unsigned int cpu)
 {
 	struct mcs_spinlock *base = per_cpu_ptr(&qnodes[0].mcs, cpu);
@@ -98,6 +124,7 @@ 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;
 }
 
 /*
@@ -197,8 +224,15 @@ 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 = (s32)jiffies;
+		/* make sure start_time != 0 iff secondary queue is not empty */
+		if (!cn->start_time)
+			cn->start_time = 1;
 	} else {
 		/* add to the tail of the secondary queue */
 		struct mcs_spinlock *tail_2nd = decode_tail(node->locked);
@@ -249,11 +283,15 @@ static __always_inline u32 cna_wait_head_or_lock(struct qspinlock *lock,
 {
 	struct cna_node *cn = (struct cna_node *)node;
 
-	/*
-	 * Try and put the time otherwise spent spin waiting on
-	 * _Q_LOCKED_PENDING_MASK to use by sorting our lists.
-	 */
-	cn->partial_order = cna_order_queue(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.
+		 */
+		cn->partial_order = cna_order_queue(node);
+	} else {
+		cn->partial_order = FLUSH_SECONDARY_QUEUE;
+	}
 
 	return 0; /* we lied; we didn't wait, go do so now */
 }
@@ -276,13 +314,29 @@ static inline void cna_lock_handoff(struct mcs_spinlock *node,
 	 */
 	WARN_ON(partial_order == LOCAL_WAITER_NOT_FOUND);
 
-	/*
-	 * We found a local waiter; reload @next in case it was changed by
-	 * cna_order_queue().
-	 */
-	next = node->next;
-	if (node->locked > 1)
-		val = node->locked;	/* preseve secondary queue */
+	if (partial_order == LOCAL_WAITER_FOUND) {
+		/*
+		 * We found a local waiter; reload @next in case it
+		 * was changed by cna_order_queue().
+		 */
+		next = node->next;
+		if (node->locked > 1) {
+			val = node->locked;     /* preseve secondary queue */
+			((struct cna_node *)next)->start_time = cn->start_time;
+		}
+	} else {
+		WARN_ON(partial_order != FLUSH_SECONDARY_QUEUE);
+		/*
+		 * We decided to flush the secondary queue;
+		 * this can only happen if that queue is not empty.
+		 */
+		WARN_ON(node->locked <= 1);
+		/*
+		 * Splice the secondary queue onto the primary queue and pass the lock
+		 * to the longest waiting remote waiter.
+		 */
+		next = cna_splice_head(NULL, 0, node, next);
+	}
 
 	arch_mcs_lock_handoff(&next->locked, val);
 }
@@ -334,3 +388,20 @@ void __init cna_configure_spin_lock_slowpath(void)
 
 	pr_info("Enabling CNA spinlock\n");
 }
+
+static int __init numa_spinlock_threshold_setup(char *str)
+{
+	int param;
+
+	if (get_option(&str, &param)) {
+		/* valid value is between 1 and 100 */
+		if (param <= 0 || param > 100)
+			return 0;
+
+		intra_node_handoff_threshold = msecs_to_jiffies(param);
+		return 1;
+	}
+
+	return 0;
+}
+__setup("numa_spinlock_threshold=", numa_spinlock_threshold_setup);
-- 
2.7.4

[PATCH v12 5/5] locking/qspinlock: Avoid moving certain threads between waiting queues in CNA

[Alex Kogan]

Prohibit moving certain threads (e.g., in irq and nmi contexts)
to the secondary queue. Those prioritized threads will always stay
in the primary queue, and so will have a shorter wait time for the lock.

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>
Reviewed-by: Steve Sistare <steven.sistare@oracle.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 kernel/locking/qspinlock_cna.h | 26 ++++++++++++++++++++------
 1 file changed, 20 insertions(+), 6 deletions(-)

diff --git a/kernel/locking/qspinlock_cna.h b/kernel/locking/qspinlock_cna.h
index d3e2754..ac3109a 100644
--- a/kernel/locking/qspinlock_cna.h
+++ b/kernel/locking/qspinlock_cna.h
@@ -4,6 +4,7 @@
 #endif
 
 #include <linux/topology.h>
+#include <linux/sched/rt.h>
 
 /*
  * Implement a NUMA-aware version of MCS (aka CNA, or compact NUMA-aware lock).
@@ -35,7 +36,8 @@
  * running on the same NUMA node. If it is not, that waiter is detached from the
  * main queue and moved into the tail of the secondary queue. This way, we
  * gradually filter the primary queue, leaving only waiters running on the same
- * preferred NUMA node.
+ * preferred NUMA node. Note that certain priortized waiters (e.g., in
+ * irq and nmi contexts) are excluded from being moved to the secondary queue.
  *
  * We change the NUMA node preference after a waiter at the head of the
  * secondary queue spins for a certain amount of time (10ms, by default).
@@ -49,6 +51,8 @@
  *          Dave Dice <dave.dice@oracle.com>
  */
 
+#define CNA_PRIORITY_NODE      0xffff
+
 struct cna_node {
 	struct mcs_spinlock	mcs;
 	u16			numa_node;
@@ -121,9 +125,10 @@ static int __init cna_init_nodes(void)
 
 static __always_inline void cna_init_node(struct mcs_spinlock *node)
 {
+	bool priority = !in_task() || irqs_disabled() || rt_task(current);
 	struct cna_node *cn = (struct cna_node *)node;
 
-	cn->numa_node = cn->real_numa_node;
+	cn->numa_node = priority ? CNA_PRIORITY_NODE : cn->real_numa_node;
 	cn->start_time = 0;
 }
 
@@ -262,11 +267,13 @@ static u32 cna_order_queue(struct mcs_spinlock *node)
 	next_numa_node = ((struct cna_node *)next)->numa_node;
 
 	if (next_numa_node != numa_node) {
-		struct mcs_spinlock *nnext = READ_ONCE(next->next);
+		if (next_numa_node != CNA_PRIORITY_NODE) {
+			struct mcs_spinlock *nnext = READ_ONCE(next->next);
 
-		if (nnext) {
-			cna_splice_next(node, next, nnext);
-			next = nnext;
+			if (nnext) {
+				cna_splice_next(node, next, nnext);
+				next = nnext;
+			}
 		}
 		/*
 		 * Inherit NUMA node id of primary queue, to maintain the
@@ -285,6 +292,13 @@ static __always_inline u32 cna_wait_head_or_lock(struct qspinlock *lock,
 
 	if (!cn->start_time || !intra_node_threshold_reached(cn)) {
 		/*
+		 * We are at the head of the wait queue, no need to use
+		 * the fake NUMA node ID.
+		 */
+		if (cn->numa_node == CNA_PRIORITY_NODE)
+			cn->numa_node = cn->real_numa_node;
+
+		/*
 		 * Try and put the time otherwise spent spin waiting on
 		 * _Q_LOCKED_PENDING_MASK to use by sorting our lists.
 		 */
-- 
2.7.4