mm: zero unavailable pages before memmap init
authorPavel Tatashin <pasha.tatashin@oracle.com>
Sat, 14 Jul 2018 13:15:07 +0000 (09:15 -0400)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 14 Jul 2018 18:02:20 +0000 (11:02 -0700)
commite181ae0c5db9544de9c53239eb22bc012ce75033
tree4af6ac8786ab6cc14c292b8dbf8aa8b75de62b30
parent2db39a2f491a48ec740e0214a7dd584eefc2137d
mm: zero unavailable pages before memmap init

We must zero struct pages for memory that is not backed by physical
memory, or kernel does not have access to.

Recently, there was a change which zeroed all memmap for all holes in
e820.  Unfortunately, it introduced a bug that is discussed here:

  https://www.spinics.net/lists/linux-mm/msg156764.html

Linus, also saw this bug on his machine, and confirmed that reverting
commit 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into
memblock.reserved") fixes the issue.

The problem is that we incorrectly zero some struct pages after they
were setup.

The fix is to zero unavailable struct pages prior to initializing of
struct pages.

A more detailed fix should come later that would avoid double zeroing
cases: one in __init_single_page(), the other one in
zero_resv_unavail().

Fixes: 124049decbb1 ("x86/e820: put !E820_TYPE_RAM regions into memblock.reserved")
Signed-off-by: Pavel Tatashin <pasha.tatashin@oracle.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/page_alloc.c