Revert "mm: use early_pfn_to_nid in page_ext_init"
authorQian Cai <cai@lca.pw>
Tue, 12 Feb 2019 23:36:03 +0000 (15:36 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 13 Feb 2019 00:33:18 +0000 (16:33 -0800)
This reverts commit fe53ca54270a ("mm: use early_pfn_to_nid in
page_ext_init").

When booting a system with "page_owner=on",

start_kernel
  page_ext_init
    invoke_init_callbacks
      init_section_page_ext
        init_page_owner
          init_early_allocated_pages
            init_zones_in_node
              init_pages_in_zone
                lookup_page_ext
                  page_to_nid

The issue here is that page_to_nid() will not work since some page flags
have no node information until later in page_alloc_init_late() due to
DEFERRED_STRUCT_PAGE_INIT.  Hence, it could trigger an out-of-bounds
access with an invalid nid.

  UBSAN: Undefined behaviour in ./include/linux/mm.h:1104:50
  index 7 is out of range for type 'zone [5]'

Also, kernel will panic since flags were poisoned earlier with,

CONFIG_DEBUG_VM_PGFLAGS=y
CONFIG_NODE_NOT_IN_PAGE_FLAGS=n

start_kernel
  setup_arch
    pagetable_init
      paging_init
        sparse_init
          sparse_init_nid
            memblock_alloc_try_nid_raw

It did not handle it well in init_pages_in_zone() which ends up calling
page_to_nid().

  page:ffffea0004200000 is uninitialized and poisoned
  raw: ffffffffffffffff ffffffffffffffff ffffffffffffffff ffffffffffffffff
  raw: ffffffffffffffff ffffffffffffffff ffffffffffffffff ffffffffffffffff
  page dumped because: VM_BUG_ON_PAGE(PagePoisoned(p))
  page_owner info is not active (free page?)
  kernel BUG at include/linux/mm.h:990!
  RIP: 0010:init_page_owner+0x486/0x520

This means that assumptions behind commit fe53ca54270a ("mm: use
early_pfn_to_nid in page_ext_init") are incomplete.  Therefore, revert
the commit for now.  A proper way to move the page_owner initialization
to sooner is to hook into memmap initialization.

Link: http://lkml.kernel.org/r/20190115202812.75820-1-cai@lca.pw
Signed-off-by: Qian Cai <cai@lca.pw>
Acked-by: Michal Hocko <mhocko@kernel.org>
Cc: Pasha Tatashin <Pavel.Tatashin@microsoft.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Yang Shi <yang.shi@linaro.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
init/main.c
mm/page_ext.c

index e2e80ca3165a8726c8f13cbd6d62dcadbefb5abc..c86a1c8f19f40be9508b7b979959bc57a47e70ce 100644 (file)
@@ -695,7 +695,6 @@ asmlinkage __visible void __init start_kernel(void)
                initrd_start = 0;
        }
 #endif
-       page_ext_init();
        kmemleak_init();
        setup_per_cpu_pageset();
        numa_policy_init();
@@ -1131,6 +1130,8 @@ static noinline void __init kernel_init_freeable(void)
        sched_init_smp();
 
        page_alloc_init_late();
+       /* Initialize page ext after all struct pages are initialized. */
+       page_ext_init();
 
        do_basic_setup();
 
index ae44f7adbe07df3e32b5092ecc74f3c28c7416c2..8c78b8d451179710f7652bc5db7160d28b4161b8 100644 (file)
@@ -398,10 +398,8 @@ void __init page_ext_init(void)
                         * We know some arch can have a nodes layout such as
                         * -------------pfn-------------->
                         * N0 | N1 | N2 | N0 | N1 | N2|....
-                        *
-                        * Take into account DEFERRED_STRUCT_PAGE_INIT.
                         */
-                       if (early_pfn_to_nid(pfn) != nid)
+                       if (pfn_to_nid(pfn) != nid)
                                continue;
                        if (init_section_page_ext(pfn, nid))
                                goto oom;