0001-runtime_context-no-rcu.patch

From 17739c97357a43001e7498f6a226c3e90af12626 Mon Sep 17 00:00:00 2001
From: Sultan Alsawaf <sultan@openresty.com>
Date: Fri, 6 Nov 2020 13:54:05 -0800
Subject: [PATCH] runtime_context: no rcu

---
 runtime/linux/runtime_context.h | 66 +++------------------------------
 1 file changed, 5 insertions(+), 61 deletions(-)

diff --git a/runtime/linux/runtime_context.h b/runtime/linux/runtime_context.h
index 41fecba81..2dddfbbff 100644
--- a/runtime/linux/runtime_context.h
+++ b/runtime/linux/runtime_context.h
@@ -15,24 +15,10 @@
 #define __rcu
 #endif
 
-static struct context __rcu *contexts[NR_CPUS] = { NULL };
+static DEFINE_PER_CPU(struct context, contexts);
 
 static int _stp_runtime_contexts_alloc(void)
 {
-	int cpu;
-
-	for_each_possible_cpu(cpu) {
-		/* Module init, so in user context, safe to use
-		 * "sleeping" allocation. */
-		struct context *c = _stp_vzalloc_node(sizeof (struct context),
-						      cpu_to_node(cpu));
-		if (c == NULL) {
-			_stp_error ("context (size %lu per cpu) allocation failed",
-				    (unsigned long) sizeof (struct context));
-			return -ENOMEM;
-		}
-		rcu_assign_pointer(contexts[cpu], c);
-	}
 	return 0;
 }
 
@@ -41,50 +27,11 @@ static int _stp_runtime_contexts_alloc(void)
  * use RCU-sched synchronization to be sure its safe to free them.  */
 static void _stp_runtime_contexts_free(void)
 {
-	// Note that 'free_contexts' is static because it is
-	// (probably) too big to fit on a kernel function's stack.
-	static struct context *free_contexts[NR_CPUS] = { NULL };
-	int cpu;
-
-	/* First, save all the pointers.  */
-	rcu_read_lock_sched();
-	for_each_possible_cpu(cpu) {
-		free_contexts[cpu] = rcu_dereference_sched(contexts[cpu]);
-	}
-	rcu_read_unlock_sched();
-
-	/* Now clear all pointers to prevent new readers.  */
-	for_each_possible_cpu(cpu) {
-		rcu_assign_pointer(contexts[cpu], NULL);
-	}
-
-	/* Sync to make sure existing readers are done.  */
-	stp_synchronize_sched();
-
-	/* Now we can actually free the contexts.  */
-	for_each_possible_cpu(cpu) {
-		struct context *c = free_contexts[cpu];
-		if (c != NULL) {
-			free_contexts[cpu] = NULL;
-			_stp_vfree(c);
-		}
-	}
 }
 
 static inline struct context * _stp_runtime_get_context(void)
 {
-	// RHBZ1788662 rcu operations are rejected in idle-cpu contexts
-	// in effect: skip probe if it's in rcu-idle state
-#if defined(STAPCONF_RCU_IS_WATCHING) || LINUX_VERSION_CODE >= KERNEL_VERSION(3,13,0) // linux commit #5c173eb8
-        if (! rcu_is_watching())
-		return 0;
-#elif LINUX_VERSION_CODE >= KERNEL_VERSION(3,3,0) // linux commit #9b2e4f18
-        if (! rcu_is_cpu_idle())
-		return 0;
-#else
-	; // XXX older kernels didn't put tracepoints in idle-cpu
-#endif
-	return rcu_dereference_sched(contexts[smp_processor_id()]);
+	return &per_cpu(contexts, smp_processor_id());
 }
 
 static struct context * _stp_runtime_entryfn_get_context(void)
@@ -93,7 +40,7 @@ static struct context * _stp_runtime_entryfn_get_context(void)
 	preempt_disable ();
 	c = _stp_runtime_get_context();
 	if (c != NULL) {
-		if (atomic_inc_return(&c->busy) == 1) {
+		if (!atomic_cmpxchg(&c->busy, 0, 1)) {
 			// NB: Notice we're not re-enabling preemption
 			// here. We exepect the calling code to call
 			// _stp_runtime_entryfn_get_context() and
@@ -101,7 +48,6 @@ static struct context * _stp_runtime_entryfn_get_context(void)
 			// pair.
 			return c;
 		}
-		atomic_dec(&c->busy);
 	}
 	preempt_enable_no_resched();
 	return NULL;
@@ -110,9 +56,7 @@ static struct context * _stp_runtime_entryfn_get_context(void)
 static inline void _stp_runtime_entryfn_put_context(struct context *c)
 {
 	if (c) {
-		if (c == _stp_runtime_get_context())
-			atomic_dec(&c->busy);
-		/* else, warn about bad state? */
+		atomic_set(&c->busy, 0);
 		preempt_enable_no_resched();
 	}
 	return;
@@ -132,7 +76,7 @@ static void _stp_runtime_context_wait(void)
 		holdon = 0;
 		rcu_read_lock_sched();
 		for_each_possible_cpu(i) {
-			struct context *c = rcu_dereference_sched(contexts[i]);
+			struct context *c = &per_cpu(contexts, i);
 			if (c != NULL
 			    && atomic_read (& c->busy)) {
 				holdon = 1;
-- 
2.29.2