powerpc/qspinlock: allow indefinite spinning on a preempted owner

Provide an option that holds off queueing indefinitely while the lock
owner is preempted. This could reduce queueing latencies for very
overcommitted vcpu situations.

This is disabled by default.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20221126095932.1234527-16-npiggin@gmail.com

arch/powerpc/lib/qspinlock.c

@@ -35,6 +35,7 @@ static int head_spins __read_mostly = (1 << 8);
 
 static bool pv_yield_owner __read_mostly = true;
 static bool pv_yield_allow_steal __read_mostly = false;
+static bool pv_spin_on_preempted_owner __read_mostly = false;
 static bool pv_yield_prev __read_mostly = true;
 static bool pv_yield_propagate_owner __read_mostly = true;
 static bool pv_prod_head __read_mostly = false;
@@ -191,11 +192,12 @@ static struct qnode *get_tail_qnode(struct qspinlock *lock, u32 val)
 	BUG();
 }
 
-/* Called inside spin_begin() */
-static __always_inline void __yield_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt, bool mustq)
+/* Called inside spin_begin(). Returns whether or not the vCPU was preempted. */
+static __always_inline bool __yield_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt, bool mustq)
 {
 	int owner;
 	u32 yield_count;
+	bool preempted = false;
 
 	BUG_ON(!(val & _Q_LOCKED_VAL));
 
@@ -213,6 +215,8 @@ static __always_inline void __yield_to_locked_owner(struct qspinlock *lock, u32
 
 	spin_end();
 
+	preempted = true;
+
 	/*
 	 * Read the lock word after sampling the yield count. On the other side
 	 * there may a wmb because the yield count update is done by the
@@ -229,29 +233,32 @@ static __always_inline void __yield_to_locked_owner(struct qspinlock *lock, u32
 		if (mustq)
 			set_mustq(lock);
 		spin_begin();
+
 		/* Don't relax if we yielded. Maybe we should? */
-		return;
+		return preempted;
 	}
 	spin_begin();
 relax:
 	spin_cpu_relax();
+
+	return preempted;
 }
 
-/* Called inside spin_begin() */
-static __always_inline void yield_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt)
+/* Called inside spin_begin(). Returns whether or not the vCPU was preempted. */
+static __always_inline bool yield_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt)
 {
-	__yield_to_locked_owner(lock, val, paravirt, false);
+	return __yield_to_locked_owner(lock, val, paravirt, false);
 }
 
-/* Called inside spin_begin() */
-static __always_inline void yield_head_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt)
+/* Called inside spin_begin(). Returns whether or not the vCPU was preempted. */
+static __always_inline bool yield_head_to_locked_owner(struct qspinlock *lock, u32 val, bool paravirt)
 {
 	bool mustq = false;
 
 	if ((val & _Q_MUST_Q_VAL) && pv_yield_allow_steal)
 		mustq = true;
 
-	__yield_to_locked_owner(lock, val, paravirt, mustq);
+	return __yield_to_locked_owner(lock, val, paravirt, mustq);
 }
 
 static __always_inline void propagate_yield_cpu(struct qnode *node, u32 val, int *set_yield_cpu, bool paravirt)
@@ -361,13 +368,16 @@ static __always_inline bool try_to_steal_lock(struct qspinlock *lock, bool parav
 	int iters = 0;
 	u32 val;
 
-	if (!steal_spins)
+	if (!steal_spins) {
+		/* XXX: should spin_on_preempted_owner do anything here? */
 		return false;
+	}
 
 	/* Attempt to steal the lock */
 	spin_begin();
-
 	do {
+		bool preempted = false;
+
 		val = READ_ONCE(lock->val);
 		if (val & _Q_MUST_Q_VAL)
 			break;
@@ -378,10 +388,23 @@ static __always_inline bool try_to_steal_lock(struct qspinlock *lock, bool parav
 				return true;
 			spin_begin();
 		} else {
-			yield_to_locked_owner(lock, val, paravirt);
+			preempted = yield_to_locked_owner(lock, val, paravirt);
 		}
 
-		iters++;
+		if (preempted) {
+			if (!pv_spin_on_preempted_owner)
+				iters++;
+			/*
+			 * pv_spin_on_preempted_owner don't increase iters
+			 * while the owner is preempted -- we won't interfere
+			 * with it by definition. This could introduce some
+			 * latency issue if we continually observe preempted
+			 * owners, but hopefully that's a rare corner case of
+			 * a badly oversubscribed system.
+			 */
+		} else {
+			iters++;
+		}
 	} while (!steal_break(val, iters, paravirt));
 
 	spin_end();
@@ -453,15 +476,22 @@ static __always_inline void queued_spin_lock_mcs_queue(struct qspinlock *lock, b
 	/* We're at the head of the waitqueue, wait for the lock. */
 	spin_begin();
 	for (;;) {
+		bool preempted;
+
 		val = READ_ONCE(lock->val);
 		if (!(val & _Q_LOCKED_VAL))
 			break;
 
 		propagate_yield_cpu(node, val, &set_yield_cpu, paravirt);
-		yield_head_to_locked_owner(lock, val, paravirt);
+		preempted = yield_head_to_locked_owner(lock, val, paravirt);
 		if (!maybe_stealers)
 			continue;
-		iters++;
+		if (preempted) {
+			if (!pv_spin_on_preempted_owner)
+				iters++;
+		} else {
+			iters++;
+		}
 
 		if (!mustq && iters >= get_head_spins(paravirt)) {
 			mustq = true;
@@ -644,6 +674,22 @@ static int pv_yield_allow_steal_get(void *data, u64 *val)
 
 DEFINE_SIMPLE_ATTRIBUTE(fops_pv_yield_allow_steal, pv_yield_allow_steal_get, pv_yield_allow_steal_set, "%llu\n");
 
+static int pv_spin_on_preempted_owner_set(void *data, u64 val)
+{
+	pv_spin_on_preempted_owner = !!val;
+
+	return 0;
+}
+
+static int pv_spin_on_preempted_owner_get(void *data, u64 *val)
+{
+	*val = pv_spin_on_preempted_owner;
+
+	return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(fops_pv_spin_on_preempted_owner, pv_spin_on_preempted_owner_get, pv_spin_on_preempted_owner_set, "%llu\n");
+
 static int pv_yield_prev_set(void *data, u64 val)
 {
 	pv_yield_prev = !!val;
@@ -700,6 +746,7 @@ static __init int spinlock_debugfs_init(void)
 	if (is_shared_processor()) {
 		debugfs_create_file("qspl_pv_yield_owner", 0600, arch_debugfs_dir, NULL, &fops_pv_yield_owner);
 		debugfs_create_file("qspl_pv_yield_allow_steal", 0600, arch_debugfs_dir, NULL, &fops_pv_yield_allow_steal);
+		debugfs_create_file("qspl_pv_spin_on_preempted_owner", 0600, arch_debugfs_dir, NULL, &fops_pv_spin_on_preempted_owner);
 		debugfs_create_file("qspl_pv_yield_prev", 0600, arch_debugfs_dir, NULL, &fops_pv_yield_prev);
 		debugfs_create_file("qspl_pv_yield_propagate_owner", 0600, arch_debugfs_dir, NULL, &fops_pv_yield_propagate_owner);
 		debugfs_create_file("qspl_pv_prod_head", 0600, arch_debugfs_dir, NULL, &fops_pv_prod_head);