Re: [PATCH 4/6] cpuset: don't allocate trial cpuset on stack

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 31 Dec 2008 16:36:41 +0800 Li Zefan <lizf@cn.fujitsu.com> wrote:

> Impact: cleanups, reduce stack usage
> 
> This patch prepares for the next patch. When we convert cpuset.cpus_allowed
> to cpumask_var_t, (trialcs = *cs) no longer works.
> 
> Another result of this patch is reducing stack usage of trialcs. sizeof(*cs)
> can be as large as 148 bytes on x86_64, so it's really not good to have it
> on stack.
> 

err, what?

> --- a/kernel/cpuset.c
> +++ b/kernel/cpuset.c
> @@ -415,6 +415,24 @@ static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
>  		is_mem_exclusive(p) <= is_mem_exclusive(q);
>  }
>  
> +/**
> + * alloc_trial_cpuset - allocate a trial cpuset
> + * @cs: the cpuset that the trial cpuset duplicates
> + */
> +static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
> +{
> +	return kmemdup(cs, sizeof(*cs), GFP_KERNEL);
> +}

We copy a cpuset by value?

> -static int update_cpumask(struct cpuset *cs, const char *buf)
> +static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
> +			  const char *buf)
>  {
>  	struct ptr_heap heap;
> -	struct cpuset trialcs;
>  	int retval;
>  	int is_load_balanced;
>  
> @@ -891,8 +909,6 @@ static int update_cpumask(struct cpuset *cs, const char *buf)
>  	if (cs == &top_cpuset)
>  		return -EACCES;
>  
> -	trialcs = *cs;

Yes, we already do.

That thing contains spinlocks and list_heads (at least), which cannot
be copied in this way.

Seems that we're doing this gross thing because it just so happens that
we only use the cpus_allowed and mems_allowed fields, and because
several of the called functions require a cpuset*, but only needed a
cpumask_t.

How perfectly beastly.