[PATCH] workqueue: Free useless memory when disabling NUMA in wq_numa_init()

[PATCH] workqueue: Free useless memory when disabling NUMA in wq_numa_init()

[Xunlei Pang]

From: Xunlei Pang <pang.xunlei@zte.com.cn>

wq_update_unbound_numa_attrs_buf and tbl will never be used
in case of invalid NUMA node mapping, so just free them.

Signed-off-by: Xunlei Pang <pang.xunlei@zte.com.cn>
---
 kernel/workqueue.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index ca71582..22d7747 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -5197,6 +5197,13 @@ static void __init wq_numa_init(void)
 		node = cpu_to_node(cpu);
 		if (WARN_ON(node == NUMA_NO_NODE)) {
 			pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu);
+
+			for_each_node(node)
+				free_cpumask_var(tbl[node]);
+			kfree(tbl);
+
+			free_workqueue_attrs(wq_update_unbound_numa_attrs_buf);
+			wq_update_unbound_numa_attrs_buf = NULL;
 			/* happens iff arch is bonkers, let's just proceed */
 			return;
 		}
-- 
1.9.1

Re: [PATCH] workqueue: Free useless memory when disabling NUMA in wq_numa_init()

[Tejun Heo]

Hello,

On Thu, Sep 17, 2015 at 10:43:55AM +0800, Xunlei Pang wrote:
> From: Xunlei Pang <pang.xunlei@zte.com.cn>
> 
> wq_update_unbound_numa_attrs_buf and tbl will never be used
> in case of invalid NUMA node mapping, so just free them.
> 
> Signed-off-by: Xunlei Pang <pang.xunlei@zte.com.cn>
> ---
>  kernel/workqueue.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
> index ca71582..22d7747 100644
> --- a/kernel/workqueue.c
> +++ b/kernel/workqueue.c
> @@ -5197,6 +5197,13 @@ static void __init wq_numa_init(void)
>  		node = cpu_to_node(cpu);
>  		if (WARN_ON(node == NUMA_NO_NODE)) {
>  			pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu);
> +
> +			for_each_node(node)
> +				free_cpumask_var(tbl[node]);
> +			kfree(tbl);

So, this is "either bios or arch code buggy but let's keep limping on
rather than crashing" path.  It doesn't matter whether there's
something lingering behind.  We wanna keep it as simple as possible.

Thanks.

-- 
tejun