Re: [PATCH 4/4] workqueue: reuse wq_update_unbound_numa_attrs_buf as temporary attrs

[Tejun Heo <tj@kernel.org>]

Hello,

On Wed, Jun 03, 2015 at 10:29:52PM +0800, Lai Jiangshan wrote:
> tmp_attrs in apply_wqattrs_prepare() is just temporary attrs, we can use
> wq_update_unbound_numa_attrs_buf for it like wq_update_unbound_numa();
> 
> The wq_update_unbound_numa_attrs_buf is renamed to wq_calc_node_attrs_buf
> since it is used both for apply_wqattrs_prepare() and wq_update_unbound_numa().
> 
> The comment for using wq_calc_node_attrs_buf in wq_update_unbound_numa()
> is also moved to the defination of the wq_calc_node_attrs_buf.
> 
> This change also avoids frequently alloc/free the tmp_attrs for every
> workqueue when the low level cpumask is being updated.
> 
> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> ---
>  kernel/workqueue.c | 31 +++++++++++++------------------
>  1 file changed, 13 insertions(+), 18 deletions(-)
> 
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index 0c2f819..4a40165 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -295,8 +295,13 @@ module_param_named(power_efficient, wq_power_efficient, bool, 0444);
>  
>  static bool wq_numa_enabled;		/* unbound NUMA affinity enabled */
>  
> -/* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */
> -static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf;
> +/*
> + * PL: resulted attrs of wq_calc_node_cpumask() for apply_wqattrs_prepare()
> + * and wq_update_unbound_numa().
> + * We don't wanna alloc/free temporary attrs for each call. Let's preallocate
> + * one with the access protection of wq_pool_mutex.
> + */
> +static struct workqueue_attrs *wq_calc_node_attrs_buf;
>  
>  static DEFINE_MUTEX(wq_pool_mutex);	/* protects pools and workqueues list */
>  static DEFINE_SPINLOCK(wq_mayday_lock);	/* protects wq->maydays list */
> @@ -3515,7 +3520,7 @@ apply_wqattrs_prepare(struct workqueue_struct *wq,
>  		      const struct workqueue_attrs *attrs)
>  {
>  	struct apply_wqattrs_ctx *ctx;
> -	struct workqueue_attrs *new_attrs, *tmp_attrs;
> +	struct workqueue_attrs *new_attrs, *tmp_attrs = wq_calc_node_attrs_buf;

Hmm... this actually bothers me a bit.  Can we please do something
like the following?

static struct workqueue_attrs *wq_attrs_shared_buf(void)
{
	lockdep_assert_held(...);
	return __wq_attrs_shared_buf;
}

-- 
tejun