Re: [PATCH] workqueue, lockdep: Fix new alloc_workqueue() error path

[Kefeng Wang]

Please ignore, no check lkml again, see https://lkml.org/lkml/2019/3/11/1274:-)

On 2019/3/18 19:58, Kefeng Wang wrote:
> If alloc_workqueue_attrs() fails, the workqueue lockdep is
> uninitialized, it will trigger a warning in err_free_wq path
> when try to unregister workqueue lockdep.
>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Bart Van Assche <bvanassche@xxxxxxx>
> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> Cc: Tejun Heo <tj@xxxxxxxxxx>
> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Fixes: 009bb421b6ce ("workqueue, lockdep: Fix an alloc_workqueue() error path")
> Signed-off-by: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
> ---
>  kernel/workqueue.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 4026d1871407..2c19f2968bc7 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -4242,7 +4242,7 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,
>  	if (flags & WQ_UNBOUND) {
>  		wq->unbound_attrs = alloc_workqueue_attrs(GFP_KERNEL);
>  		if (!wq->unbound_attrs)
> -			goto err_free_wq;
> +			goto err_unbound_attrs;
>  	}
>  
>  	va_start(args, max_active);
> @@ -4296,6 +4296,7 @@ struct workqueue_struct *alloc_workqueue(const char *fmt,
>  	wq_unregister_lockdep(wq);
>  	wq_free_lockdep(wq);
>  	free_workqueue_attrs(wq->unbound_attrs);
> +err_unbound_attrs:
>  	kfree(wq);
>  	return NULL;
>  err_destroy: