On Tue, Oct 19, 2021 at 10:56:48AM +0200, Peter Zijlstra wrote:

On Mon, Oct 18, 2021 at 11:47:32PM -0400, Qian Cai wrote:

quoted

Peter, any thoughts? I did confirm that reverting the commit fixed the issue.

On 10/13/2021 10:32 AM, Qian Cai wrote:

quoted

On 9/29/2021 11:17 AM, Peter Zijlstra wrote:

quoted

--- a/kernel/smp.c
+++ b/kernel/smp.c

@@ -1170,14 +1170,14 @@ void wake_up_all_idle_cpus(void)
 {
 	int cpu;
 
-	preempt_disable();
+	cpus_read_lock();
 	for_each_online_cpu(cpu) {
-		if (cpu == smp_processor_id())
+		if (cpu == raw_smp_processor_id())
 			continue;
 
 		wake_up_if_idle(cpu);
 	}
-	preempt_enable();
+	cpus_read_unlock();

Right, so yesterday I thought: YW2KGrvvv/vSA+97@hirez.programming.kicks-ass.net
but today I might have another idea, lemme go prod at this a bit more.

Here, hows this then?

---

diff --git a/kernel/smp.c b/kernel/smp.c
index ad0b68a3a3d3..61ddc7a3bafa 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c

@@ -1170,14 +1170,12 @@ void wake_up_all_idle_cpus(void)
 {
 	int cpu;
 
-	cpus_read_lock();
-	for_each_online_cpu(cpu) {
-		if (cpu == raw_smp_processor_id())
-			continue;
-
-		wake_up_if_idle(cpu);
+	for_each_cpu(cpu) {
+		preempt_disable();
+		if (cpu != smp_processor_id() && cpu_online(cpu))
+			wake_up_if_idle(cpu);
+		preempt_enable();
 	}
-	cpus_read_unlock();
 }
 EXPORT_SYMBOL_GPL(wake_up_all_idle_cpus);