Re: kworker with empty task->cpus_allowed (was Re: [v4.12-rc1 regression] mount ext4 fs results in kernel crash on PPC64le host)

[Michael Ellerman]

Eryu Guan <eg...@redhat.com> writes:
> On Fri, Jun 30, 2017 at 08:07:02PM +1000, Michael Ellerman wrote:
>> 
>> Can you try this patch and see if it changes anything? (with the debug
>> still applied).
>
> This patch fixes the crash for me. After appliying this patch (with all
> other debug patches still applied), kernel didn't print any warnings or
> calltraces or debug messages.

OK. It's not meant to fix it :)

I can't form any connection between your bisection result and that
patch, nothing is making any sense TBH.

What hardware are you on? And are you doing CPU hotplug or anything like that?

Can you back out the last patch I sent and try this?

cheers


diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index c74bf39ef764..8ec3841f9689 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -3338,6 +3338,8 @@ static struct worker_pool *get_unbound_pool(const struct 
workqueue_attrs *attrs)
 
        lockdep_assert_held(&wq_pool_mutex);
 
+       WARN_ON(cpumask_empty(attrs->cpumask));
+
        /* do we already have a matching pool? */
        hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) {
                if (wqattrs_equal(pool->attrs, attrs)) {
@@ -3366,6 +3368,8 @@ static struct worker_pool *get_unbound_pool(const struct 
workqueue_attrs *attrs)
        copy_workqueue_attrs(pool->attrs, attrs);
        pool->node = target_node;
 
+       WARN_ON(cpumask_empty(pool->attrs->cpumask));
+
        /*
         * no_numa isn't a worker_pool attribute, always clear it.  See
         * 'struct workqueue_attrs' comments for detail.
@@ -5494,6 +5498,7 @@ static void __init wq_numa_init(void)
 
        for_each_possible_cpu(cpu) {
                node = cpu_to_node(cpu);
+               printk("%s: setting cpu %d on node %d present? %d\n", __func__, 
cpu, node, cpu_present(cpu));
                if (WARN_ON(node == NUMA_NO_NODE)) {
                        pr_warn("workqueue: NUMA node mapping not available for 
cpu%d, disabling NUMA support\n", cpu);
                        /* happens iff arch is bonkers, let's just proceed */
@@ -5502,6 +5507,16 @@ static void __init wq_numa_init(void)
                cpumask_set_cpu(cpu, tbl[node]);
        }
 
+       for_each_possible_cpu(cpu) {
+               struct worker_pool *pool;
+
+               for_each_cpu_worker_pool(pool, cpu) {
+                       if (cpumask_empty(pool->attrs->cpumask))
+                               printk("%s: cpumask EMPTY! for pool %p on cpu 
%d\n", __func__, pool, cpu);
+                       printk("%s: pool %p on cpu %d node = %d\n", __func__, 
pool, cpu, pool->node);
+               }
+       }
+
        wq_numa_possible_cpumask = tbl;
        wq_numa_enabled = true;
 }