powerpc/smp: Create idle threads on demand and properly reset them

Instead of creating idle threads at boot for all possible CPUs, we
create them on demand, like x86 or ARM, and we properly call init_idle
to re-initialize an idle thread when a CPU was unplugged and is now
re-plugged.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

1 files changed, 78 insertions, 22 deletions

diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
index d7f8cc1..54faff9 100644
--- a/arch/powerpc/kernel/smp.c
+++ b/arch/powerpc/kernel/smp.c
@@ -57,6 +57,25 @@
 #define DBG(fmt...)
 #endif
 
+
+/* Store all idle threads, this can be reused instead of creating
+* a new thread. Also avoids complicated thread destroy functionality
+* for idle threads.
+*/
+#ifdef CONFIG_HOTPLUG_CPU
+/*
+ * Needed only for CONFIG_HOTPLUG_CPU because __cpuinitdata is
+ * removed after init for !CONFIG_HOTPLUG_CPU.
+ */
+static DEFINE_PER_CPU(struct task_struct *, idle_thread_array);
+#define get_idle_for_cpu(x)      (per_cpu(idle_thread_array, x))
+#define set_idle_for_cpu(x, p)   (per_cpu(idle_thread_array, x) = (p))
+#else
+static struct task_struct *idle_thread_array[NR_CPUS] __cpuinitdata ;
+#define get_idle_for_cpu(x)      (idle_thread_array[(x)])
+#define set_idle_for_cpu(x, p)   (idle_thread_array[(x)] = (p))
+#endif
+
 struct thread_info *secondary_ti;
 
 DEFINE_PER_CPU(cpumask_var_t, cpu_sibling_map);
@@ -238,23 +257,6 @@ static void __devinit smp_store_cpu_info(int id)
 	per_cpu(cpu_pvr, id) = mfspr(SPRN_PVR);
 }
 
-static void __init smp_create_idle(unsigned int cpu)
-{
-	struct task_struct *p;
-
-	/* create a process for the processor */
-	p = fork_idle(cpu);
-	if (IS_ERR(p))
-		panic("failed fork for CPU %u: %li", cpu, PTR_ERR(p));
-#ifdef CONFIG_PPC64
-	paca[cpu].__current = p;
-	paca[cpu].kstack = (unsigned long) task_thread_info(p)
-		+ THREAD_SIZE - STACK_FRAME_OVERHEAD;
-#endif
-	current_set[cpu] = task_thread_info(p);
-	task_thread_info(p)->cpu = cpu;
-}
-
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
 	unsigned int cpu;
@@ -288,10 +290,6 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
 			max_cpus = NR_CPUS;
 	else
 		max_cpus = 1;
-
-	for_each_possible_cpu(cpu)
-		if (cpu != boot_cpuid)
-			smp_create_idle(cpu);
 }
 
 void __devinit smp_prepare_boot_cpu(void)
@@ -355,9 +353,62 @@ void generic_set_cpu_dead(unsigned int cpu)
 }
 #endif
 
+struct create_idle {
+	struct work_struct work;
+	struct task_struct *idle;
+	struct completion done;
+	int cpu;
+};
+
+static void __cpuinit do_fork_idle(struct work_struct *work)
+{
+	struct create_idle *c_idle =
+		container_of(work, struct create_idle, work);
+
+	c_idle->idle = fork_idle(c_idle->cpu);
+	complete(&c_idle->done);
+}
+
+static int __cpuinit create_idle(unsigned int cpu)
+{
+	struct thread_info *ti;
+	struct create_idle c_idle = {
+		.cpu	= cpu,
+		.done	= COMPLETION_INITIALIZER_ONSTACK(c_idle.done),
+	};
+	INIT_WORK_ONSTACK(&c_idle.work, do_fork_idle);
+
+	c_idle.idle = get_idle_for_cpu(cpu);
+
+	/* We can't use kernel_thread since we must avoid to
+	 * reschedule the child. We use a workqueue because
+	 * we want to fork from a kernel thread, not whatever
+	 * userspace process happens to be trying to online us.
+	 */
+	if (!c_idle.idle) {
+		schedule_work(&c_idle.work);
+		wait_for_completion(&c_idle.done);
+	} else
+		init_idle(c_idle.idle, cpu);
+	if (IS_ERR(c_idle.idle)) {		
+		pr_err("Failed fork for CPU %u: %li", cpu, PTR_ERR(c_idle.idle));
+		return PTR_ERR(c_idle.idle);
+	}
+	ti = task_thread_info(c_idle.idle);
+
+#ifdef CONFIG_PPC64
+	paca[cpu].__current = c_idle.idle;
+	paca[cpu].kstack = (unsigned long)ti + THREAD_SIZE - STACK_FRAME_OVERHEAD;
+#endif
+	ti->cpu = cpu;
+	current_set[cpu] = ti;
+
+	return 0;
+}
+
 int __cpuinit __cpu_up(unsigned int cpu)
 {
-	int c;
+	int rc, c;
 
 	secondary_ti = current_set[cpu];
 
@@ -365,6 +416,11 @@ int __cpuinit __cpu_up(unsigned int cpu)
 	    (smp_ops->cpu_bootable && !smp_ops->cpu_bootable(cpu)))
 		return -EINVAL;
 
+	/* Make sure we have an idle thread */
+	rc = create_idle(cpu);
+	if (rc)
+		return rc;
+
 	/* Make sure callin-map entry is 0 (can be leftover a CPU
 	 * hotplug
 	 */