target/linux/generic/patches-4.4/103-Fix-alloc_node_mem_map-with-ARCH_PFN_OFFSET-calcu.patch

   1 From: Tobias Wolf <dev-NTEO@vplace.de>
   2 Date: Wed, 30 Nov 2016 09:16:41 +0100
   3 Subject: [PATCH] mm: Fix alloc_node_mem_map with ARCH_PFN_OFFSET
   4  calculation
   5
   6 Dear folks,
   7
   8 An rt288x (ralink) based router (Belkin F5D8235 v1) does not boot with any
   9 kernel beyond version 4.3 resulting in:
  10
  11 BUG: Bad page state in process swapper  pfn:086ac
  12
  13 bisect resulted in:
  14
  15 a1c34a3bf00af2cede839879502e12dc68491ad5 is the first bad commit
  16 commit a1c34a3bf00af2cede839879502e12dc68491ad5
  17 Author: Laura Abbott <laura@labbott.name>
  18 Date:   Thu Nov 5 18:48:46 2015 -0800
  19
  20     mm: Don't offset memmap for flatmem
  21
  22     Srinivas Kandagatla reported bad page messages when trying to remove the
  23     bottom 2MB on an ARM based IFC6410 board
  24
  25       BUG: Bad page state in process swapper  pfn:fffa8
  26       page:ef7fb500 count:0 mapcount:0 mapping:  (null) index:0x0
  27       flags: 0x96640253(locked|error|dirty|active|arch_1|reclaim|mlocked)
  28       page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
  29       bad because of flags:
  30       flags: 0x200041(locked|active|mlocked)
  31       Modules linked in:
  32       CPU: 0 PID: 0 Comm: swapper Not tainted 3.19.0-rc3-00007-g412f9ba-dirty
  33 #816
  34       Hardware name: Qualcomm (Flattened Device Tree)
  35         unwind_backtrace
  36         show_stack
  37         dump_stack
  38         bad_page
  39         free_pages_prepare
  40         free_hot_cold_page
  41         __free_pages
  42         free_highmem_page
  43         mem_init
  44         start_kernel
  45       Disabling lock debugging due to kernel taint
  46     [...]
  47 :040000 040000 2de013c372345fd471cd58f0553c9b38b0ef1cc4
  48 0a8156f848733dfa21e16c196dfb6c0a76290709 M      mm
  49
  50 This fix for ARM does not account ARCH_PFN_OFFSET for mem_map as later used by
  51 page_to_pfn anymore.
  52
  53 The following output was generated with two hacked in printk statements:
  54
  55 printk("before %p vs. %p or %p\n", mem_map, mem_map - offset, mem_map -
  56 (pgdat->node_start_pfn - ARCH_PFN_OFFSET));
  57                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  58                         mem_map -= offset + (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
  59 printk("after %p\n", mem_map);
  60
  61 Output:
  62
  63 [    0.000000] before 8861b280 vs. 8861b280 or 8851b280
  64 [    0.000000] after 8851b280
  65
  66 As seen in the first line mem_map with subtraction of offset does not equal the
  67 mem_map after subtraction of ARCH_PFN_OFFSET.
  68
  69 After adding the offset of ARCH_PFN_OFFSET as well to mem_map as the
  70 previously calculated offset is zero for the named platform it is able to boot
  71 4.4 and 4.9-rc7 again.
  72
  73 Signed-off-by: Tobias Wolf <dev-NTEO@vplace.de>
  74 ---
  75
  76 --- a/mm/page_alloc.c
  77 +++ b/mm/page_alloc.c
  78 @@ -5357,7 +5357,7 @@ static void __init_refok alloc_node_mem_
  79                 mem_map = NODE_DATA(0)->node_mem_map;
  80  #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
  81                 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
  82 -                       mem_map -= offset;
  83 +                       mem_map -= offset + (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
  84  #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
  85         }
  86  #endif