Re: [PATCH 3/4] mm: introduce memmap_alloc() to unify memory map allocation

From: Mike Rapoport
Date: Thu Jul 15 2021 - 02:10:42 EST


On Wed, Jul 14, 2021 at 03:32:08PM -0700, Andrew Morton wrote:
> On Wed, 14 Jul 2021 15:37:38 +0300 Mike Rapoport <rppt@xxxxxxxxxx> wrote:
>
> > From: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> >
> > There are several places that allocate memory for the memory map:
> > alloc_node_mem_map() for FLATMEM, sparse_buffer_init() and
> > __populate_section_memmap() for SPARSEMEM.
> >
> > The memory allocated in the FLATMEM case is zeroed and it is never
> > poisoned, regardless of CONFIG_PAGE_POISON setting.
> >
> > The memory allocated in the SPARSEMEM cases is not zeroed and it is
> > implicitly poisoned inside memblock if CONFIG_PAGE_POISON is set.
> >
> > Introduce memmap_alloc() wrapper for memblock allocators that will be used
> > for both FLATMEM and SPARSEMEM cases and will makei memory map zeroing and
> > poisoning consistent for different memory models.
> >
> > ...
> >
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -6730,6 +6730,26 @@ static void __init memmap_init(void)
> > init_unavailable_range(hole_pfn, end_pfn, zone_id, nid);
> > }
> >
> > +void __init *memmap_alloc(phys_addr_t size, phys_addr_t align,
> > + phys_addr_t min_addr, int nid, bool exact_nid)
> > +{
> > + void *ptr;
> > +
> > + if (exact_nid)
> > + ptr = memblock_alloc_exact_nid_raw(size, align, min_addr,
> > + MEMBLOCK_ALLOC_ACCESSIBLE,
> > + nid);
> > + else
> > + ptr = memblock_alloc_try_nid_raw(size, align, min_addr,
> > + MEMBLOCK_ALLOC_ACCESSIBLE,
> > + nid);
> > +
> > + if (ptr && size > 0)
> > + page_init_poison(ptr, size);
> > +
> > + return ptr;
> > +}
> > +
> > static int zone_batchsize(struct zone *zone)
> > {
> > #ifdef CONFIG_MMU
> > @@ -7501,8 +7521,8 @@ static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
> > end = pgdat_end_pfn(pgdat);
> > end = ALIGN(end, MAX_ORDER_NR_PAGES);
> > size = (end - start) * sizeof(struct page);
> > - map = memblock_alloc_node(size, SMP_CACHE_BYTES,
> > - pgdat->node_id);
> > + map = memmap_alloc(size, SMP_CACHE_BYTES, MEMBLOCK_LOW_LIMIT,
> > + pgdat->node_id, false);
>
> Mostly offtopic, but... Why is alloc_node_mem_map() marked __ref?

Once free_area_init_node() was __meminit, I stopped digging at that point.

> afaict it can be __init?

Yes.

--
Sincerely yours,
Mike.