]> git.hungrycats.org Git - linux/commit
mm: don't do zero_resv_unavail if memmap is not allocated
authorPavel Tatashin <pasha.tatashin@oracle.com>
Mon, 16 Jul 2018 15:16:30 +0000 (11:16 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 18 Jul 2018 05:56:38 +0000 (07:56 +0200)
commit27d8b7daf73827e8263ab0b740da6ebdda9f9d56
tree2aa9e2251daaa81d699f48f118abfc865346ed4e
parent73bc05003e9b1aebd1c473c0b42e6e4d08d209d5
mm: don't do zero_resv_unavail if memmap is not allocated

commit d1b47a7c9efcf3c3384b70f6e3c8f1423b44d8c7 upstream.

Moving zero_resv_unavail before memmap_init_zone(), caused a regression on
x86-32.

The cause is that we access struct pages before they are allocated when
CONFIG_FLAT_NODE_MEM_MAP is used.

free_area_init_nodes()
  zero_resv_unavail()
    mm_zero_struct_page(pfn_to_page(pfn)); <- struct page is not alloced
  free_area_init_node()
    if CONFIG_FLAT_NODE_MEM_MAP
      alloc_node_mem_map()
        memblock_virt_alloc_node_nopanic() <- struct page alloced here

On the other hand memblock_virt_alloc_node_nopanic() zeroes all the memory
that it returns, so we do not need to do zero_resv_unavail() here.

Fixes: e181ae0c5db9 ("mm: zero unavailable pages before memmap init")
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Tested-by: Matt Hart <matt@mattface.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
include/linux/mm.h
mm/page_alloc.c