rcu: Add rcu_normal kernel parameter to suppress expediting

Although expedited grace periods can be quite useful, and although their
OS jitter has been greatly reduced, they can still pose problems for
extreme real-time workloads.  This commit therefore adds a rcu_normal
kernel boot parameter (which can also be manipulated via sysfs)
to suppress expedited grace periods, that is, to treat requests for
expedited grace periods as if they were requests for normal grace periods.
If both rcu_expedited and rcu_normal are specified, rcu_normal wins.
This means that if you are relying on expedited grace periods to speed up
boot, you will want to specify rcu_expedited on the kernel command line,
and then specify rcu_normal via sysfs once boot completes.

Signed-off-by: Paul E. McKenney <[email protected]>

Documentation/kernel-parameters.txt

@@ -3296,18 +3296,27 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
 	rcutorture.verbose= [KNL]
 			Enable additional printk() statements.
 
+	rcupdate.rcu_cpu_stall_suppress= [KNL]
+			Suppress RCU CPU stall warning messages.
+
+	rcupdate.rcu_cpu_stall_timeout= [KNL]
+			Set timeout for RCU CPU stall warning messages.
+
 	rcupdate.rcu_expedited= [KNL]
 			Use expedited grace-period primitives, for
 			example, synchronize_rcu_expedited() instead
 			of synchronize_rcu().  This reduces latency,
 			but can increase CPU utilization, degrade
 			real-time latency, and degrade energy efficiency.
 
-	rcupdate.rcu_cpu_stall_suppress= [KNL]
-			Suppress RCU CPU stall warning messages.
-
-	rcupdate.rcu_cpu_stall_timeout= [KNL]
-			Set timeout for RCU CPU stall warning messages.
+	rcupdate.rcu_normal= [KNL]
+			Use only normal grace-period primitives,
+			for example, synchronize_rcu() instead of
+			synchronize_rcu_expedited().  This improves
+			real-time latency, CPU utilization, and energy
+			efficiency, but can expose users to increased
+			grace-period latency.  This parameter overrides
+			rcupdate.rcu_expedited.
 
 	rcupdate.rcu_task_stall_timeout= [KNL]
 			Set timeout in jiffies for RCU task stall warning

include/linux/rcupdate.h

@@ -49,9 +49,14 @@
 #include <asm/barrier.h>
 
 extern int rcu_expedited; /* for sysctl */
+extern int rcu_normal;    /* also for sysctl */
 
 #ifdef CONFIG_TINY_RCU
 /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */
+static inline bool rcu_gp_is_normal(void)  /* Internal RCU use. */
+{
+	return true;
+}
 static inline bool rcu_gp_is_expedited(void)  /* Internal RCU use. */
 {
 	return false;
@@ -65,6 +70,7 @@ static inline void rcu_unexpedite_gp(void)
 {
 }
 #else /* #ifdef CONFIG_TINY_RCU */
+bool rcu_gp_is_normal(void);     /* Internal RCU use. */
 bool rcu_gp_is_expedited(void);  /* Internal RCU use. */
 void rcu_expedite_gp(void);
 void rcu_unexpedite_gp(void);

kernel/ksysfs.c

@@ -20,7 +20,7 @@
 #include <linux/capability.h>
 #include <linux/compiler.h>
 
-#include <linux/rcupdate.h>	/* rcu_expedited */
+#include <linux/rcupdate.h>	/* rcu_expedited and rcu_normal */
 
 #define KERNEL_ATTR_RO(_name) \
 static struct kobj_attribute _name##_attr = __ATTR_RO(_name)
@@ -148,7 +148,7 @@ int rcu_expedited;
 static ssize_t rcu_expedited_show(struct kobject *kobj,
 				  struct kobj_attribute *attr, char *buf)
 {
-	return sprintf(buf, "%d\n", rcu_expedited);
+	return sprintf(buf, "%d\n", READ_ONCE(rcu_expedited));
 }
 static ssize_t rcu_expedited_store(struct kobject *kobj,
 				   struct kobj_attribute *attr,
@@ -161,6 +161,23 @@ static ssize_t rcu_expedited_store(struct kobject *kobj,
 }
 KERNEL_ATTR_RW(rcu_expedited);
 
+int rcu_normal;
+static ssize_t rcu_normal_show(struct kobject *kobj,
+			       struct kobj_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", READ_ONCE(rcu_normal));
+}
+static ssize_t rcu_normal_store(struct kobject *kobj,
+				struct kobj_attribute *attr,
+				const char *buf, size_t count)
+{
+	if (kstrtoint(buf, 0, &rcu_normal))
+		return -EINVAL;
+
+	return count;
+}
+KERNEL_ATTR_RW(rcu_normal);
+
 /*
  * Make /sys/kernel/notes give the raw contents of our kernel .notes section.
  */
@@ -203,6 +220,7 @@ static struct attribute * kernel_attrs[] = {
 	&vmcoreinfo_attr.attr,
 #endif
 	&rcu_expedited_attr.attr,
+	&rcu_normal_attr.attr,
 	NULL
 };
 

kernel/rcu/srcu.c

@@ -489,7 +489,7 @@ static void __synchronize_srcu(struct srcu_struct *sp, int trycount)
  */
 void synchronize_srcu(struct srcu_struct *sp)
 {
-	__synchronize_srcu(sp, rcu_gp_is_expedited()
+	__synchronize_srcu(sp, (rcu_gp_is_expedited() && !rcu_gp_is_normal())
 			   ? SYNCHRONIZE_SRCU_EXP_TRYCOUNT
 			   : SYNCHRONIZE_SRCU_TRYCOUNT);
 }

kernel/rcu/tree.c

@@ -3841,6 +3841,12 @@ void synchronize_sched_expedited(void)
 	if (rcu_blocking_is_gp())
 		return;
 
+	/* If expedited grace periods are prohibited, fall back to normal. */
+	if (rcu_gp_is_normal()) {
+		wait_rcu_gp(call_rcu_sched);
+		return;
+	}
+
 	/* Take a snapshot of the sequence number.  */
 	s = rcu_exp_gp_seq_snap(rsp);
 

kernel/rcu/tree_plugin.h

@@ -746,6 +746,12 @@ void synchronize_rcu_expedited(void)
 	struct rcu_state *rsp = rcu_state_p;
 	unsigned long s;
 
+	/* If expedited grace periods are prohibited, fall back to normal. */
+	if (rcu_gp_is_normal()) {
+		wait_rcu_gp(call_rcu);
+		return;
+	}
+
 	s = rcu_exp_gp_seq_snap(rsp);
 
 	rnp_unlock = exp_funnel_lock(rsp, s);

kernel/rcu/update.c

@@ -61,6 +61,7 @@ MODULE_ALIAS("rcupdate");
 #define MODULE_PARAM_PREFIX "rcupdate."
 
 module_param(rcu_expedited, int, 0);
+module_param(rcu_normal, int, 0);
 
 #if defined(CONFIG_DEBUG_LOCK_ALLOC) && defined(CONFIG_PREEMPT_COUNT)
 /**
@@ -113,6 +114,17 @@ EXPORT_SYMBOL(rcu_read_lock_sched_held);
 
 #ifndef CONFIG_TINY_RCU
 
+/*
+ * Should expedited grace-period primitives always fall back to their
+ * non-expedited counterparts?  Intended for use within RCU.  Note
+ * that if the user specifies both rcu_expedited and rcu_normal, then
+ * rcu_normal wins.
+ */
+bool rcu_gp_is_normal(void)
+{
+	return READ_ONCE(rcu_normal);
+}
+
 static atomic_t rcu_expedited_nesting =
 	ATOMIC_INIT(IS_ENABLED(CONFIG_RCU_EXPEDITE_BOOT) ? 1 : 0);