[PATCH 2/3] stop_machine: add timeout for child thread deployment

[Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>]

If stopmachine() invoked while one of onlined cpu is locked up
by some reason, stopmachine cannot finish its work because the
locked cpu cannot stop.

This patch allows stopmachine to return -EBUSY if any of
kstopmachine's child threads cannot start running on its target
cpu.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
---
 kernel/stop_machine.c |   40 +++++++++++++++++++++++++++++++++++++++-
 1 files changed, 39 insertions(+), 1 deletions(-)

Index: GIT-torvalds/kernel/stop_machine.c
===================================================================
--- GIT-torvalds.orig/kernel/stop_machine.c	2008-04-29 00:29:20.000000000 +0900
+++ GIT-torvalds/kernel/stop_machine.c	2008-04-29 00:31:55.000000000 +0900
@@ -29,6 +29,9 @@
 static enum stopmachine_state stopmachine_state;
 static unsigned int stopmachine_num_threads;
 static atomic_t stopmachine_thread_ack;
+static atomic_t stopmachine_busy_exit;
+
+static unsigned long stopmachine_timeout = 5; /* secs, arbitrary */
 
 static int stopmachine(void *cpu)
 {
@@ -42,6 +45,7 @@
 	if (stopmachine_state == STOPMACHINE_EXIT)
 		goto exit;
 
+	/* If target cpu is on fire, this call can stuck */
 	set_cpus_allowed_ptr(current, &cpumask_of_cpu((int)(long)cpu));
 
 	/* Ack: we arrived */
@@ -83,6 +87,12 @@
 	if (prepared)
 		preempt_enable();
 
+	if (atomic_read(&stopmachine_busy_exit)) {
+		atomic_dec(&stopmachine_busy_exit);
+		printk(KERN_INFO "stopmachine: cpu#%d is not busy now.\n",
+			(int)(long)cpu);
+	}
+
 	return 0;
 }
 
@@ -99,6 +109,15 @@
 static int stop_machine(void)
 {
 	int i, ret = 0;
+	unsigned long limit;
+
+	if (atomic_read(&stopmachine_busy_exit)) {
+		/*
+		 * previous try was timeout, and still there is a unreachable
+		 * cpu and abandoned child.
+		 */
+		return -EBUSY;
+	}
 
 	atomic_set(&stopmachine_thread_ack, 0);
 	stopmachine_num_threads = 0;
@@ -113,10 +132,15 @@
 		stopmachine_num_threads++;
 	}
 
+	limit = jiffies + msecs_to_jiffies(stopmachine_timeout * MSEC_PER_SEC);
+
 	/* Wait for them all to come to life on the target. */
 	stopmachine_state = STOPMACHINE_DEPLOY;
 	while (atomic_read(&stopmachine_thread_ack) != stopmachine_num_threads)
-		yield();
+		if (time_is_after_jiffies(limit))
+			yield();
+		else
+			goto deploy_timeout;
 
 	/* Now they are all started, make them hold the CPUs, ready. */
 	preempt_disable();
@@ -129,6 +153,20 @@
 
 	return 0;
 
+deploy_timeout:
+	printk(KERN_CRIT "stopmachine: Failed to stop machine in time(%lds). "
+		"Are there any CPUs on file?\n", stopmachine_timeout);
+
+	/* defer exit check to the beginning of next try. */
+	atomic_set(&stopmachine_busy_exit, stopmachine_num_threads);
+
+	printk(KERN_INFO "stopmachine: cpu#%d is initiator of failed stop.\n",
+			raw_smp_processor_id());
+	smp_wmb();
+	stopmachine_state = STOPMACHINE_EXIT;
+
+	return -EBUSY;
+
 exit_threads:
 	/* Wait for them all to exit, since stop is canceled */
 	stopmachine_set_state(STOPMACHINE_EXIT);