Re: [PATCH] mm/mm_init: remove unnecessary initialization of pgdat->per_cpu_nodestats

[Mike Rapoport <rppt@kernel.org>]

On Wed, Jul 29, 2026 at 01:27:45PM -0700, Andrew Morton wrote:
> On Thu, 30 Jul 2026 00:51:15 +0900 Sang-Heon Jeon <ekffu200098@gmail.com> wrote:
> 
> > free_area_init_node() sets pgdat->per_cpu_nodestats to NULL and later
> > calls free_area_init_core(), which unconditionally overwrites it with
> > &boot_nodestats. Nothing reads the field in between, so the store has
> > no effect.
> > 
> > Remove unnecessary initialization. No functional change.
> > 
> > ...
> >
> > --- a/mm/mm_init.c
> > +++ b/mm/mm_init.c
> > @@ -1722,7 +1722,6 @@ static void __init free_area_init_node(int nid)
> >  
> >  	pgdat->node_id = nid;
> >  	pgdat->node_start_pfn = start_pfn;
> > -	pgdat->per_cpu_nodestats = NULL;
> >  
> >  	if (start_pfn != end_pfn) {
> >  		pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
> 
> Looks OK, thanks.  Although this statement is "free" in __init code and
> its removal does make the code that bit harder to follow.

Heh, it's already not that easy to follow :)

We initialize a few fields in pgdat, then setup node and its zone sizes,
then return to initialization of the rest of pgdat internals.

It feels like we need a larger cleanup there, and this patch wouldn't
really move a needle.
 
> Possibly *pgdat was zeroed out anyway, I got bored figuring that out. 
> The memblock_alloc_range_nid() kerneldoc is silent on the state of the
> returned memory but I spy a kzalloc_node() in there.

memblock_alloc_range_nid() does not zero the memory, unless inappropriately
called after slab_is_available(). But alloc_node_data() that allocates
pgdat's does :)

-- 
Sincerely yours,
Mike.