Re: [PATCH v2] rcu: Add per-CPU rcuc task info to RCU CPU stall warnings

From: Ammar Faizi
Date: Mon Jan 24 2022 - 18:09:15 EST


On 1/24/22 11:42 PM, Paul E. McKenney wrote:
On Mon, Jan 24, 2022 at 05:38:21PM +0700, Ammar Faizi wrote:
[snip...]
FWIW, this one makes more sense:
```
static void rcuc_kthread_dump(struct rcu_data *rdp)
{
int cpu;
unsigned long j;
struct task_struct *rcuc;

if (!rcu_is_rcuc_kthread_starving(rdp, &j))
return;

rcuc = rdp->rcu_cpu_kthread_task;
if (!rcuc)
return;

pr_err("%s kthread starved for %ld jiffies, stack dump:\n", rcuc->comm, j);

Thank you for looking this over and for the great feedback, Ammar!

I am also wondering why the above message should be printed when the
corresponding CPU is offline or idle. Why not move the above pr_err()
line down to replace the pr_err() line below?

Thanx, Paul

Hi Paul, Thank you for the review. Agree with that.
Hopefully this one looks better (untested):
```
static void rcuc_kthread_dump(struct rcu_data *rdp)
{
int cpu;
unsigned long j;
struct task_struct *rcuc;

rcuc = rdp->rcu_cpu_kthread_task;
if (!rcuc)
return;

cpu = task_cpu(rcuc);
if (cpu_is_offline(cpu) || idle_cpu(cpu))
return;

if (!rcu_is_rcuc_kthread_starving(rdp, &j))
return;

pr_err("%s kthread starved for %ld jiffies\n", rcuc->comm, j);
sched_show_task(rcuc);
if (!trigger_single_cpu_backtrace(cpu))
dump_cpu_task(cpu);
}
```

Recall that dump_cpu_task looks like this:
```
void dump_cpu_task(int cpu)
{
pr_info("Task dump for CPU %d:\n", cpu);
sched_show_task(cpu_curr(cpu));
}
```
which already tells us it's a dump, so "stack dump" in the pr_err()
can be omitted. Any comment, Zqiang?

--
Ammar Faizi