Re: [PATCH] mm: Remove VM_BUG_ON in __alloc_pages_node

[Michal Hocko]

On Wed 05-06-19 11:32:29, Bharath Vedartham wrote:
> In __alloc_pages_node, there is a VM_BUG_ON on the condition (nid < 0 ||
> nid >= MAX_NUMNODES). Remove this VM_BUG_ON and add a VM_WARN_ON, if the
> condition fails and fail the allocation if an invalid NUMA node id is
> passed to __alloc_pages_node.

What is the motivation of the patch? VM_BUG_ON is not enabled by default
and your patch adds a branch to a really hot path. Why is this an
improvement for something that shouldn't happen in the first place?

> 
> The check (nid < 0 || nid >= MAX_NUMNODES) also considers NUMA_NO_NODE
> as an invalid nid, but the caller of __alloc_pages_node is assumed to
> have checked for the case where nid == NUMA_NO_NODE.
> 
> Signed-off-by: Bharath Vedartham <linux.bhar@xxxxxxxxx>
> ---
>  include/linux/gfp.h | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/gfp.h b/include/linux/gfp.h
> index 5f5e25f..075bdaf 100644
> --- a/include/linux/gfp.h
> +++ b/include/linux/gfp.h
> @@ -480,7 +480,11 @@ __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
>  static inline struct page *
>  __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
>  {
> -	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
> +	if (nid < 0 || nid >= MAX_NUMNODES) {
> +		VM_WARN_ON(nid < 0 || nid >= MAX_NUMNODES);
> +		return NULL; 
> +	}
> +
>  	VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
>  
>  	return __alloc_pages(gfp_mask, order, nid);
> -- 
> 2.7.4
> 

-- 
Michal Hocko
SUSE Labs