Re: [PATCH] 4.4.86-rt99: fix sync breakage between nr_cpus_allowed and cpus_allowed

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 20 Nov 2017 23:02:07 -0500
Steven Rostedt <rostedt@goodmis.org> wrote:


> Ideally, I would like to stay close to what upstream -rt does. Would
> you be able to backport the 4.11-rt patch?
> 
> I'm currently working on releasing 4.9-rt and 4.4-rt with the latest
> backports. I could easily add this one too.

Speaking of which. I just backported this patch to 4.4-rt. Is this what
you are talking about?

-- Steve

>From 1dc89be37874bfc7bb4a0ea7c45492d7db39f62b Mon Sep 17 00:00:00 2001
From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date: Mon, 19 Jun 2017 09:55:47 +0200
Subject: [PATCH] sched/migrate disable: handle updated task-mask mg-dis
 section

If task's cpumask changes while in the task is in a migrate_disable()
section then we don't react on it after a migrate_enable(). It matters
however if current CPU is no longer part of the cpumask. We also miss
the ->set_cpus_allowed() callback.
This patch fixes it by setting task->migrate_disable_update once we this
"delayed" hook.
This bug was introduced while fixing unrelated issue in
migrate_disable() in v4.4-rt3 (update_migrate_disable() got removed
during that).

Cc: stable-rt@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
---
 include/linux/sched.h |    1 
 kernel/sched/core.c   |   59 ++++++++++++++++++++++++++++++++++++++++++++------
 2 files changed, 54 insertions(+), 6 deletions(-)

Index: stable-rt.git/include/linux/sched.h
===================================================================
--- stable-rt.git.orig/include/linux/sched.h	2017-11-20 23:43:24.214077537 -0500
+++ stable-rt.git/include/linux/sched.h	2017-11-20 23:43:24.154079278 -0500
@@ -1438,6 +1438,7 @@ struct task_struct {
 	unsigned int policy;
 #ifdef CONFIG_PREEMPT_RT_FULL
 	int migrate_disable;
+	int migrate_disable_update;
 # ifdef CONFIG_SCHED_DEBUG
 	int migrate_disable_atomic;
 # endif
Index: stable-rt.git/kernel/sched/core.c
===================================================================
--- stable-rt.git.orig/kernel/sched/core.c	2017-11-20 23:43:24.214077537 -0500
+++ stable-rt.git/kernel/sched/core.c	2017-11-20 23:56:05.071687323 -0500
@@ -1212,18 +1212,14 @@ void set_cpus_allowed_common(struct task
 	p->nr_cpus_allowed = cpumask_weight(new_mask);
 }
 
-void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
+static void __do_set_cpus_allowed_tail(struct task_struct *p,
+				       const struct cpumask *new_mask)
 {
 	struct rq *rq = task_rq(p);
 	bool queued, running;
 
 	lockdep_assert_held(&p->pi_lock);
 
-	if (__migrate_disabled(p)) {
-		cpumask_copy(&p->cpus_allowed, new_mask);
-		return;
-	}
-
 	queued = task_on_rq_queued(p);
 	running = task_current(rq, p);
 
@@ -1246,6 +1242,20 @@ void do_set_cpus_allowed(struct task_str
 		enqueue_task(rq, p, ENQUEUE_RESTORE);
 }
 
+void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask)
+{
+	if (__migrate_disabled(p)) {
+		lockdep_assert_held(&p->pi_lock);
+
+		cpumask_copy(&p->cpus_allowed, new_mask);
+#if defined(CONFIG_PREEMPT_RT_FULL) && defined(CONFIG_SMP)
+		p->migrate_disable_update = 1;
+#endif
+		return;
+	}
+	__do_set_cpus_allowed_tail(p, new_mask);
+}
+
 static DEFINE_PER_CPU(struct cpumask, sched_cpumasks);
 static DEFINE_MUTEX(sched_down_mutex);
 static cpumask_t sched_down_cpumask;
@@ -3231,6 +3241,43 @@ void migrate_enable(void)
 	 */
 	p->migrate_disable = 0;
 
+	if (p->migrate_disable_update) {
+		unsigned long flags;
+		struct rq *rq;
+
+		rq = task_rq_lock(p, &flags);
+		update_rq_clock(rq);
+
+		__do_set_cpus_allowed_tail(p, &p->cpus_allowed);
+		task_rq_unlock(rq, p, &flags);
+
+		p->migrate_disable_update = 0;
+
+		WARN_ON(smp_processor_id() != task_cpu(p));
+		if (!cpumask_test_cpu(task_cpu(p), &p->cpus_allowed)) {
+			const struct cpumask *cpu_valid_mask = cpu_active_mask;
+			struct migration_arg arg;
+			unsigned int dest_cpu;
+
+			if (p->flags & PF_KTHREAD) {
+				/*
+				 * Kernel threads are allowed on online && !active CPUs
+				 */
+				cpu_valid_mask = cpu_online_mask;
+			}
+			dest_cpu = cpumask_any_and(cpu_valid_mask, &p->cpus_allowed);
+			arg.task = p;
+			arg.dest_cpu = dest_cpu;
+
+			unpin_current_cpu();
+			preempt_lazy_enable();
+			preempt_enable();
+			stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg);
+			tlb_migrate_finish(p->mm);
+			return;
+		}
+	}
+
 	unpin_current_cpu();
 	preempt_enable();
 	preempt_lazy_enable();