releases/4.19.53/x86-mm-kaslr-compute-the-size-of-the-vmemmap-section-properly.patch

   1 From 00e5a2bbcc31d5fea853f8daeba0f06c1c88c3ff Mon Sep 17 00:00:00 2001
   2 From: Baoquan He <bhe@redhat.com>
   3 Date: Thu, 23 May 2019 10:57:44 +0800
   4 Subject: x86/mm/KASLR: Compute the size of the vmemmap section properly
   5
   6 From: Baoquan He <bhe@redhat.com>
   7
   8 commit 00e5a2bbcc31d5fea853f8daeba0f06c1c88c3ff upstream.
   9
  10 The size of the vmemmap section is hardcoded to 1 TB to support the
  11 maximum amount of system RAM in 4-level paging mode - 64 TB.
  12
  13 However, 1 TB is not enough for vmemmap in 5-level paging mode. Assuming
  14 the size of struct page is 64 Bytes, to support 4 PB system RAM in 5-level,
  15 64 TB of vmemmap area is needed:
  16
  17   4 * 1000^5 PB / 4096 bytes page size * 64 bytes per page struct / 1000^4 TB = 62.5 TB.
  18
  19 This hardcoding may cause vmemmap to corrupt the following
  20 cpu_entry_area section, if KASLR puts vmemmap very close to it and the
  21 actual vmemmap size is bigger than 1 TB.
  22
  23 So calculate the actual size of the vmemmap region needed and then align
  24 it up to 1 TB boundary.
  25
  26 In 4-level paging mode it is always 1 TB. In 5-level it's adjusted on
  27 demand. The current code reserves 0.5 PB for vmemmap on 5-level. With
  28 this change, the space can be saved and thus used to increase entropy
  29 for the randomization.
  30
  31  [ bp: Spell out how the 64 TB needed for vmemmap is computed and massage commit
  32    message. ]
  33
  34 Fixes: eedb92abb9bb ("x86/mm: Make virtual memory layout dynamic for CONFIG_X86_5LEVEL=y")
  35 Signed-off-by: Baoquan He <bhe@redhat.com>
  36 Signed-off-by: Borislav Petkov <bp@suse.de>
  37 Reviewed-by: Kees Cook <keescook@chromium.org>
  38 Acked-by: Kirill A. Shutemov <kirill@linux.intel.com>
  39 Cc: Andy Lutomirski <luto@kernel.org>
  40 Cc: Dave Hansen <dave.hansen@linux.intel.com>
  41 Cc: "H. Peter Anvin" <hpa@zytor.com>
  42 Cc: Ingo Molnar <mingo@kernel.org>
  43 Cc: kirill.shutemov@linux.intel.com
  44 Cc: Peter Zijlstra <peterz@infradead.org>
  45 Cc: stable <stable@vger.kernel.org>
  46 Cc: Thomas Gleixner <tglx@linutronix.de>
  47 Cc: x86-ml <x86@kernel.org>
  48 Link: https://lkml.kernel.org/r/20190523025744.3756-1-bhe@redhat.com
  49 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  50
  51 ---
  52  arch/x86/mm/kaslr.c |   11 ++++++++++-
  53  1 file changed, 10 insertions(+), 1 deletion(-)
  54
  55 --- a/arch/x86/mm/kaslr.c
  56 +++ b/arch/x86/mm/kaslr.c
  57 @@ -51,7 +51,7 @@ static __initdata struct kaslr_memory_re
  58  } kaslr_regions[] = {
  59         { &page_offset_base, 0 },
  60         { &vmalloc_base, 0 },
  61 -       { &vmemmap_base, 1 },
  62 +       { &vmemmap_base, 0 },
  63  };
  64
  65  /* Get size in bytes used by the memory region */
  66 @@ -77,6 +77,7 @@ void __init kernel_randomize_memory(void
  67         unsigned long rand, memory_tb;
  68         struct rnd_state rand_state;
  69         unsigned long remain_entropy;
  70 +       unsigned long vmemmap_size;
  71
  72         vaddr_start = pgtable_l5_enabled() ? __PAGE_OFFSET_BASE_L5 : __PAGE_OFFSET_BASE_L4;
  73         vaddr = vaddr_start;
  74 @@ -108,6 +109,14 @@ void __init kernel_randomize_memory(void
  75         if (memory_tb < kaslr_regions[0].size_tb)
  76                 kaslr_regions[0].size_tb = memory_tb;
  77
  78 +       /*
  79 +        * Calculate the vmemmap region size in TBs, aligned to a TB
  80 +        * boundary.
  81 +        */
  82 +       vmemmap_size = (kaslr_regions[0].size_tb << (TB_SHIFT - PAGE_SHIFT)) *
  83 +                       sizeof(struct page);
  84 +       kaslr_regions[2].size_tb = DIV_ROUND_UP(vmemmap_size, 1UL << TB_SHIFT);
  85 +
  86         /* Calculate entropy available between regions */
  87         remain_entropy = vaddr_end - vaddr_start;
  88         for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)