Re: [PATCH] mm/isolation: Remove redundant pfn_valid_within() in __first_valid_page()

[Oscar Salvador]

On Thu, Mar 21, 2019 at 09:43:15AM +0530, Anshuman Khandual wrote:
> pfn_valid_within() calls pfn_valid() when CONFIG_HOLES_IN_ZONE making it
> redundant for both definitions (w/wo CONFIG_MEMORY_HOTPLUG) of the helper
> pfn_to_online_page() which either calls pfn_valid() or pfn_valid_within().
> pfn_valid_within() being 1 when !CONFIG_HOLES_IN_ZONE is irrelevant either
> way. This does not change functionality.
> 
> Fixes: 2ce13640b3f4 ("mm: __first_valid_page skip over offline pages")
> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>

About the "Fixes:" tag issue, I agree with Michal that the code is not
really broken, but perhaps "suboptimal" depending on how much can affect
performance on those systems where pfn_valid_within() is more complicated than
simple returning true.

I see that on arm64, that calls memblock_is_map_memory()->memblock_search(),
to trigger a search for the region containing the address, so I guess it
is an expensive operation.

Depending on how much time we can shave, it might be worth to have the tag
Fixes, but the removal of the code is fine anyway, so:

Reviewed-by: Oscar Salvador <osalvador@xxxxxxx>

> ---
>  mm/page_isolation.c | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/mm/page_isolation.c b/mm/page_isolation.c
> index ce323e56b34d..d9b02bb13d60 100644
> --- a/mm/page_isolation.c
> +++ b/mm/page_isolation.c
> @@ -150,8 +150,6 @@ __first_valid_page(unsigned long pfn, unsigned long nr_pages)
>  	for (i = 0; i < nr_pages; i++) {
>  		struct page *page;
>  
> -		if (!pfn_valid_within(pfn + i))
> -			continue;
>  		page = pfn_to_online_page(pfn + i);
>  		if (!page)
>  			continue;
> -- 
> 2.20.1
> 

-- 
Oscar Salvador
SUSE L3