patches/kernel/0190-x86-mm-pti-Disable-global-pages-if-PAGE_TABLE_ISOLAT.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: Dave Hansen <dave.hansen@linux.intel.com>
   3 Date: Mon, 4 Dec 2017 15:07:34 +0100
   4 Subject: [PATCH] x86/mm/pti: Disable global pages if PAGE_TABLE_ISOLATION=y
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 CVE-2017-5754
  10
  11 Global pages stay in the TLB across context switches.  Since all contexts
  12 share the same kernel mapping, these mappings are marked as global pages
  13 so kernel entries in the TLB are not flushed out on a context switch.
  14
  15 But, even having these entries in the TLB opens up something that an
  16 attacker can use, such as the double-page-fault attack:
  17
  18    http://www.ieee-security.org/TC/SP2013/papers/4977a191.pdf
  19
  20 That means that even when PAGE_TABLE_ISOLATION switches page tables
  21 on return to user space the global pages would stay in the TLB cache.
  22
  23 Disable global pages so that kernel TLB entries can be flushed before
  24 returning to user space. This way, all accesses to kernel addresses from
  25 userspace result in a TLB miss independent of the existence of a kernel
  26 mapping.
  27
  28 Suppress global pages via the __supported_pte_mask. The user space
  29 mappings set PAGE_GLOBAL for the minimal kernel mappings which are
  30 required for entry/exit. These mappings are set up manually so the
  31 filtering does not take place.
  32
  33 [ The __supported_pte_mask simplification was written by Thomas Gleixner. ]
  34 Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
  35 Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
  36 Reviewed-by: Borislav Petkov <bp@suse.de>
  37 Cc: Andy Lutomirski <luto@kernel.org>
  38 Cc: Boris Ostrovsky <boris.ostrovsky@oracle.com>
  39 Cc: Borislav Petkov <bp@alien8.de>
  40 Cc: Brian Gerst <brgerst@gmail.com>
  41 Cc: David Laight <David.Laight@aculab.com>
  42 Cc: Denys Vlasenko <dvlasenk@redhat.com>
  43 Cc: Eduardo Valentin <eduval@amazon.com>
  44 Cc: Greg KH <gregkh@linuxfoundation.org>
  45 Cc: H. Peter Anvin <hpa@zytor.com>
  46 Cc: Josh Poimboeuf <jpoimboe@redhat.com>
  47 Cc: Juergen Gross <jgross@suse.com>
  48 Cc: Linus Torvalds <torvalds@linux-foundation.org>
  49 Cc: Peter Zijlstra <peterz@infradead.org>
  50 Cc: Will Deacon <will.deacon@arm.com>
  51 Cc: aliguori@amazon.com
  52 Cc: daniel.gruss@iaik.tugraz.at
  53 Cc: hughd@google.com
  54 Cc: keescook@google.com
  55 Cc: linux-mm@kvack.org
  56 Signed-off-by: Ingo Molnar <mingo@kernel.org>
  57 (cherry picked from commit c313ec66317d421fb5768d78c56abed2dc862264)
  58 Signed-off-by: Andy Whitcroft <apw@canonical.com>
  59 Signed-off-by: Kleber Sacilotto de Souza <kleber.souza@canonical.com>
  60 (cherry picked from commit ace78e99d765da1e59f6b151adac6c360c67af7d)
  61 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  62 ---
  63  arch/x86/mm/init.c | 12 +++++++++---
  64  1 file changed, 9 insertions(+), 3 deletions(-)
  65
  66 diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
  67 index a22c2b95e513..020223420308 100644
  68 --- a/arch/x86/mm/init.c
  69 +++ b/arch/x86/mm/init.c
  70 @@ -161,6 +161,12 @@ struct map_range {
  71
  72  static int page_size_mask;
  73
  74 +static void enable_global_pages(void)
  75 +{
  76 +       if (!static_cpu_has(X86_FEATURE_PTI))
  77 +               __supported_pte_mask |= _PAGE_GLOBAL;
  78 +}
  79 +
  80  static void __init probe_page_size_mask(void)
  81  {
  82         /*
  83 @@ -179,11 +185,11 @@ static void __init probe_page_size_mask(void)
  84                 cr4_set_bits_and_update_boot(X86_CR4_PSE);
  85
  86         /* Enable PGE if available */
  87 +       __supported_pte_mask &= ~_PAGE_GLOBAL;
  88         if (boot_cpu_has(X86_FEATURE_PGE)) {
  89                 cr4_set_bits_and_update_boot(X86_CR4_PGE);
  90 -               __supported_pte_mask |= _PAGE_GLOBAL;
  91 -       } else
  92 -               __supported_pte_mask &= ~_PAGE_GLOBAL;
  93 +               enable_global_pages();
  94 +       }
  95
  96         /* Enable 1 GB linear kernel mappings if available: */
  97         if (direct_gbpages && boot_cpu_has(X86_FEATURE_GBPAGES)) {
  98 --
  99 2.14.2
 100