Re: [PATCH 08/13] workqueue: add lock_pool_executing_work()

[Lai Jiangshan <eag0628@gmail.com>]

于 2013/2/5 5:34, Tejun Heo 写道:
> Hello,
>
> On Fri, Feb 01, 2013 at 02:41:31AM +0800, Lai Jiangshan wrote:
>> +static struct worker_pool *lock_pool_executing_work(struct work_struct *work,
>> +						    struct worker **worker)
>> +{
>> +	unsigned long pool_id = offq_work_pool_id(work);
>> +	struct worker_pool *pool;
>> +	struct worker *exec;
>> +
>> +	if (pool_id == WORK_OFFQ_POOL_NONE)
>> +		return NULL;
>> +
>> +	pool = worker_pool_by_id(pool_id);
>> +	if (!pool)
>> +		return NULL;
>> +
>> +	spin_lock(&pool->lock);
>> +	exec = find_worker_executing_work(pool, work);
>> +	if (exec) {
>> +		BUG_ON(pool != exec->pool);
>> +		*worker = exec;
>> +		return pool;
>> +	}
>> +	spin_unlock(&pool->lock);
>> +
>> +	return NULL;
>> +}
>
> So, if a work item is queued on the same CPU and it isn't being
> executed, it will lock, look up the hash, unlock and then lock again?
> If this is something improved by later patch, please explain so.
> There gotta be a better way to do this, right?
>

The caller can't call this function if the work is on queue.
the caller should call it only when CWQ bit is not set.

it is commentted "lock the pool a given offq work is running on",
sorry, comment is too short.