Re: [PATCH] tracing/osnoise: fix potential deadlock in cpu hotplug

On Wed, 25 Mar 2026 10:25:42 +0800 (CST)
[off-list ref] wrote:

quoted

On Tue, 24 Mar 2026 15:06:16 +0800 (CST)
[off-list ref] wrote:
 

quoted

From: luohaiyang10243395 <redacted>

The following sequence may leads deadlock in cpu hotplug:

  CPU0                        |  CPU1
                              |  schedule_work_on
                              |
  _cpu_down//set CPU1 offline |
  cpus_write_lock             |
                              |  osnoise_hotplug_workfn
                              |    mutex_lock(&interface_lock);
                              |    cpus_read_lock();  //wait cpu_hotplug_lock
                              |
                              |  cpuhp/1
                              |    osnoise_cpu_die
                              |      kthread_stop
                              |        wait_for_completion //wait osnoise/1 exit
                              |
                              |  osnoise/1
                              |    osnoise_sleep
                              |      mutex_lock(&interface_lock); //deadlock

Fix by swap the order of cpus_read_lock() and mutex_lock(&interface_lock).  

So the deadlock is due to the "wait_for_completion"?  

The osnoise_cpu_init callback returns directly, which may allow another CPU offline task to run, 
the offline task holds the cpu_hotplug_lock while waiting for the osnoise task to exit. 
osnoise_hotplug_workfn may acquire interface_lock first, causing the offline task to be blocked. 
This is an ABBA deadlock.

Right, as I said, it is due to the "wait_for_completion" and not due to two
different locks. One is waiting for the osnoise task to exit (the
"wait_for_completion") but the osnoise task is blocked on the interface_lock().

Better to show it as:


    task1		task2		task3
    -----		-----		-----

 mutex_lock(&interface_lock)

		    [CPU GOING OFFLINE]

		    cpus_write_lock();
		    osnoise_cpu_die();
		      kthread_stop(task3);
		        wait_for_completion();

				      osnoise_sleep();
				        mutex_lock(&interface_lock);

 cpus_read_lock();

 [DEAD LOCK]

quoted

How did you find this bug? Inspection, AI, triggered?

Thanks,

-- Steve  

We run autotests on kernel-6.6, report following hung task warning, and we think the same issue exists
in linux-stable.

Thanks. It's usually good to state how a bug was discovered when fixing it.

Could you send a v2 with an updated change log?

-- Steve