Re: [PATCH] mm/page_alloc: fix boot hang in memmap_init_zone

[Michal Hocko]

On Tue 13-03-18 23:42:40, Daniel Vacek wrote:
> On some architectures (reported on arm64) commit 864b75f9d6b01 ("mm/page_alloc: fix memmap_init_zone pageblock alignment")
> causes a boot hang. This patch fixes the hang making sure the alignment
> never steps back.

I am sorry to be complaining again, but the code is so obscure that I
would _really_ appreciate some more information about what is going
on here. memblock_next_valid_pfn will most likely return a pfn within
the same memblock and the alignment will move it before the old pfn
which is not valid - so the block has some holes. Is that correct?
If yes then please put it into the changelog. Maybe reuse data provided
by Arnd http://lkml.kernel.org/r/20180314134431.13241-1-ard.biesheuvel@xxxxxxxxxx
 
> Link: http://lkml.kernel.org/r/0485727b2e82da7efbce5f6ba42524b429d0391a.1520011945.git.neelx@xxxxxxxxxx
> Fixes: 864b75f9d6b01 ("mm/page_alloc: fix memmap_init_zone pageblock alignment")
> Signed-off-by: Daniel Vacek <neelx@xxxxxxxxxx>
> Tested-by: Sudeep Holla <sudeep.holla@xxxxxxx>
> Tested-by: Naresh Kamboju <naresh.kamboju@xxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
> Cc: Michal Hocko <mhocko@xxxxxxxx>
> Cc: Paul Burton <paul.burton@xxxxxxxxxx>
> Cc: Pavel Tatashin <pasha.tatashin@xxxxxxxxxx>
> Cc: Vlastimil Babka <vbabka@xxxxxxx>
> Cc: <stable@xxxxxxxxxxxxxxx>
> ---
>  mm/page_alloc.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 3d974cb2a1a1..e033a6895c6f 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -5364,9 +5364,14 @@ void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
>  			 * is not. move_freepages_block() can shift ahead of
>  			 * the valid region but still depends on correct page
>  			 * metadata.
> +			 * Also make sure we never step back.
>  			 */
> -			pfn = (memblock_next_valid_pfn(pfn, end_pfn) &
> +			unsigned long next_pfn;
> +
> +			next_pfn = (memblock_next_valid_pfn(pfn, end_pfn) &
>  					~(pageblock_nr_pages-1)) - 1;
> +			if (next_pfn > pfn)
> +				pfn = next_pfn;
>  #endif
>  			continue;
>  		}
> -- 
> 2.16.2
> 

-- 
Michal Hocko
SUSE Labs