Re: [PATCH 2/3] mm/memblock: repeat setting reserved region nid if array is doubled

[Mike Rapoport <rppt@kernel.org>]

On Fri, Mar 14, 2025 at 02:03:51AM +0000, Wei Yang wrote:
> On Thu, Mar 13, 2025 at 05:07:59PM +0200, Mike Rapoport wrote:
> >> Commit 61167ad5fecd ("mm: pass nid to reserve_bootmem_region()") introduce
> >> a way to set nid to all reserved region.
> >> 
> >> But there is a corner case it will leave some region with invalid nid.
> >> When memblock_set_node() doubles the array of memblock.reserved, it may
> >> lead to a new reserved region before current position. The new region
> >> will be left with an invalid node id.
> >> 
> >> Repeat the process when detecting it.
> >> 
> >> Fixes: 61167ad5fecd ("mm: pass nid to reserve_bootmem_region()")
> >> Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
> >> CC: Mike Rapoport <rppt@kernel.org>
> >> CC: Yajun Deng <yajun.deng@linux.dev>
> >> CC: <stable@vger.kernel.org>
> >> ---
> >>  mm/memblock.c | 12 ++++++++++++
> >>  1 file changed, 12 insertions(+)
> >> 
> >> diff --git a/mm/memblock.c b/mm/memblock.c
> >> index 85442f1b7f14..302dd7bc622d 100644
> >> --- a/mm/memblock.c
> >> +++ b/mm/memblock.c
> >> @@ -2184,7 +2184,10 @@ static void __init memmap_init_reserved_pages(void)
> >>  	 * set nid on all reserved pages and also treat struct
> >>  	 * pages for the NOMAP regions as PageReserved
> >>  	 */
> >> +repeat:
> >>  	for_each_mem_region(region) {
> >> +		unsigned long max = memblock.reserved.max;
> >> +
> >>  		nid = memblock_get_region_node(region);
> >>  		start = region->base;
> >>  		end = start + region->size;
> >> @@ -2193,6 +2196,15 @@ static void __init memmap_init_reserved_pages(void)
> >>  			reserve_bootmem_region(start, end, nid);
> >>  
> >>  		memblock_set_node(start, region->size, &memblock.reserved, nid);
> >> +
> >> +		/*
> >> +		 * 'max' is changed means memblock.reserved has been doubled
> >> +		 * its array, which may result a new reserved region before
> >> +		 * current 'start'. Now we should repeat the procedure to set
> >> +		 * its node id.
> >> +		 */
> >> +		if (max != memblock.reserved.max)
> >> +			goto repeat;
> >
> >This check can be moved outside the loop, can't it?
> >
> 
> We can. You mean something like this?

Yes, something like this.
 
> diff --git a/mm/memblock.c b/mm/memblock.c
> index 85442f1b7f14..67fd1695cce4 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -2179,11 +2179,14 @@ static void __init memmap_init_reserved_pages(void)
>  	struct memblock_region *region;
>  	phys_addr_t start, end;
>  	int nid;
> +	unsigned long max;

maybe max_reserved?
>  
>  	/*
>  	 * set nid on all reserved pages and also treat struct
>  	 * pages for the NOMAP regions as PageReserved
>  	 */
> +repeat:
> +	max = memblock.reserved.max;
>  	for_each_mem_region(region) {
>  		nid = memblock_get_region_node(region);
>  		start = region->base;
> @@ -2194,6 +2197,13 @@ static void __init memmap_init_reserved_pages(void)
>  
>  		memblock_set_node(start, region->size, &memblock.reserved, nid);
>  	}
> +	/*
> +	 * 'max' is changed means memblock.reserved has been doubled its
> +	 * array, which may result a new reserved region before current
> +	 * 'start'. Now we should repeat the procedure to set its node id.
> +	 */
> +	if (max != memblock.reserved.max)
> +		goto repeat;
>  
>  	/*
>  	 * initialize struct pages for reserved regions that don't have
> -- 
> Wei Yang
> Help you, Help me

-- 
Sincerely yours,
Mike.