Re: [PATCH] mm/sparse: Consistently do not zero memmap

[Michal Hocko]

On Wed 30-10-19 13:53:52, Pavel Tatashin wrote:
> On Wed, Oct 30, 2019 at 1:31 PM Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> >
> > On Wed 30-10-19 12:53:41, Pavel Tatashin wrote:
[...]
> > > Yes, PMD_SIZE should be the alignment here. It just does not make
> > > sense to align to size.
> >
> > What about this? It still aligns to the size but that should be
> > correctly done to the section size level.
> >
> > diff --git a/mm/sparse.c b/mm/sparse.c
> > index 72f010d9bff5..ab1e6175ac9a 100644
> > --- a/mm/sparse.c
> > +++ b/mm/sparse.c
> > @@ -456,8 +456,7 @@ struct page __init *__populate_section_memmap(unsigned long pfn,
> >         if (map)
> >                 return map;
> >
> > -       map = memblock_alloc_try_nid(size,
> > -                                         PAGE_SIZE, addr,
> > +       map = memblock_alloc_try_nid(size, size, addr,
> >                                           MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> >         if (!map)
> >                 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa\n",
> > @@ -474,8 +473,13 @@ static void __init sparse_buffer_init(unsigned long size, int nid)
> >  {
> >         phys_addr_t addr = __pa(MAX_DMA_ADDRESS);
> >         WARN_ON(sparsemap_buf); /* forgot to call sparse_buffer_fini()? */
> > +       /*
> > +        * Pre-allocated buffer is mainly used by __populate_section_memmap
> > +        * and we want it to be properly aligned to the section size - this is
> > +        * especially the case for VMEMMAP which maps memmap to PMDs
> > +        */
> >         sparsemap_buf =
> > -               memblock_alloc_try_nid_raw(size, PAGE_SIZE,
> > +               memblock_alloc_try_nid_raw(size, section_map_size(),
> >                                                 addr,
> >                                                 MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> >         sparsemap_buf_end = sparsemap_buf + size;
> 
> This looks good, I think we should also change alignment in fallback
> of vmemmap_alloc_block() to be
> section_map_size().
> 
> +++ b/mm/sparse-vmemmap.c
> @@ -65,9 +65,10 @@ void * __meminit vmemmap_alloc_block(unsigned long
> size, int node)
>                         warned = true;
>                 }
>                 return NULL;
> -       } else
> -               return __earlyonly_bootmem_alloc(node, size, size,
> +       } else {
> +               return __earlyonly_bootmem_alloc(node, size, section_map_size(),
>                                 __pa(MAX_DMA_ADDRESS));
> +       }
>  }

Are you sure? Doesn't this provide the proper alignement already? Most
callers do PAGE_SIZE vmemmap_populate_hugepages PMD_SIZE so the
resulting alignment looks good to me.
-- 
Michal Hocko
SUSE Labs