Re: [RFC][PATCH v2 1/2] printk: Make printk() completely async

[Sergey Senozhatsky]

On (03/05/16 19:55), Sergey Senozhatsky wrote:
[..]
> +static int __init init_printk_workqueue(void)
> +{
> +	if (printk_sync)
> +		return 0;
> +
> +	printk_wq = alloc_workqueue("printk_wq", WQ_MEM_RECLAIM, 0);
> +	/*
> +	 * Fallback to one of system-wide workqueues if we failed to
> +	 * allocate a dedicated one. console_unlock() may take some
> +	 * time to flush all of the messages, so we better use
> +	 * system_long_wq.
> +	 */
> +	if (!printk_wq) {
> +		pr_err("printk: Fallback to system_long workqueue\n");
> +		printk_wq = system_long_wq;
> +	}
> +
> +	return 0;
> +}
> +early_initcall(init_printk_workqueue);
  ^^^^^^
better be late_initcall(). init_workqueues(), which creates pwq_cache kmem_cache,
is early initcall.

===8<===8<===8<====