Re: [linux-pm] [PATCH] Workqueue freezer support.

[Pavel Machek <pavel@ucw.cz>]

Hi!

> This patch implements freezer support for workqueues. The current
> refrigerator implementation makes all workqueues NOFREEZE, regardless of
> whether they need to be or not.
> 
> While this doesn't appear to have caused any problems with swsusp (ie
> Pavel's version) to date, this is no guarantee for the future.
> Furthermore, it seems better to me to treat kernel and userspace threads
> consistently, and it also enables us to err on the side of caution by
> default with new workqueues.
> 
> Queues can be made unfreezable via the new kthread_nonfreeze_run,
> create_nofreeze_workqueue and create_nofreeze_singlethread_workqueue
> calls, which take the same parameters as kthread_run, create_workqueue
> and create_singlethread_workqueue respectively. Existing call syntax is
> unchanged and the vast majority of current workqueue calls are therefore
> unaffected.
> 
> As far as Suspend2 goes, I don't rate this as critical. May save your
> hard disk partition one day, but that depends upon what workqueues get
> implemented in the future, what out of tree ones do and whether I've
> missed good rationale for having nofreeze on existing in tree instances.
> If you must flame me, call me overly careful :>.

> @@ -151,6 +158,20 @@ struct task_struct *kthread_create(int (
>  
>  	return create.result;
>  }
> +
> +struct task_struct *kthread_create(int (*threadfn)(void *data),
> +				   void *data,
> +				   const char namefmt[], ...)
> +{
> +	char result[TASK_COMM_LEN];
> +
> +	va_list args;
> +	va_start(args, namefmt);
> +	vsnprintf(result, TASK_COMM_LEN, namefmt, args);
> +	va_end(args);
> +	return _kthread_create(threadfn, data, 0, result);
> +}
> +

This is slightly ugly and uses lot of stack. Otherwise patch looks
okay. If you want me to apply it, be sure to put me into To: or at
least Cc:. Or perhaps you want to just mail it to akpm, noting that I
acked it (if you do something with the char result[] :-).

								Pavel
-- 
teflon -- maybe it is a trademark, but it should not be.