Re: [PATCH v2 2/2] Do not change worker's running cpu in cmci_rediscover().

[Tang Chen]

On 10/20/2012 01:21 AM, Luck, Tony wrote:
> > In this case, the following BUG_ON in try_to_wake_up_local() will be triggered:
> > BUG_ON(rq != this_rq());
>
> Logically this looks OK - what is the test case to trigger this?  I've done a moderate
> amount of testing of cpu online/offline while injecting corrected errors (when testing
> the CMCI storm patches) ... but didn't see this problem.

Hi Tony, Borislav,

Here is my case.

I have 2 nodes, node0 and node1. node1 could be hotpluged.
node0 has cpu0 ~ cpu15, node1 has cpu16 ~ cpu31.

I online all the cpus on node1, and hot-remove node1 directly.

When this problem is triggered, current is a work thread.

For example: cpu20 is dying. current is on cpu21, it migrates
itself to cpu22.

Assume current is process1, and it is a work thread.

       cpu21                        cpu22
p1:
....
cmci_rediscover()
|-set_cpus_allowed_ptr()
  |-stop_one_cpu()
    |-create a work to excute migration_cpu_stop()
      |-wait_for_completion()
        |-wait_for_common()
          |-might_sleep()
Here, p1 gives up cpu21.

The work starts:
migration_cpu_stop()
|-migrate p1 to cpu22

                               On cpu22, p1 wakes up:
                               p1:
                               In wait_for_common()
                                  |-do_wait_for_common()
                                    |-schedule_timeout()
                                      |-schedule()
                                        |-__schedule()
                                          |-try_to_wake_up_local()
                                            |-wq_worker_sleeping()
                                            |-BUG_ON(rq != this_rq())

On cpu22, wq_worker_sleeping() uses p1's worker_pool to find a worker
to go on to execute p1. But p1's worker_pool is on cpu21, and p1 is now
on cpu22. So the BUG_ON(rq != this_rq()) is triggered.

Thanks. :)

> -Tony

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/