mm: make CONFIG_DEFERRED_STRUCT_PAGE_INIT depends on !FLATMEM explicitly
authorYang Shi <yang.shi@linaro.org>
Thu, 26 May 2016 22:16:08 +0000 (15:16 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 26 May 2016 22:35:44 +0000 (15:35 -0700)
Per the suggestion from Michal Hocko [1], DEFERRED_STRUCT_PAGE_INIT
requires some ordering wrt other initialization operations, e.g.
page_ext_init has to happen after the whole memmap is initialized
properly.

For SPARSEMEM this requires to wait for page_alloc_init_late.  Other
memory models (e.g.  flatmem) might have different initialization
layouts (page_ext_init_flatmem).  Currently DEFERRED_STRUCT_PAGE_INIT
depends on MEMORY_HOTPLUG which in turn

depends on SPARSEMEM || X86_64_ACPI_NUMA
depends on ARCH_ENABLE_MEMORY_HOTPLUG

and X86_64_ACPI_NUMA depends on NUMA which in turn disable FLATMEM
memory model:

config ARCH_FLATMEM_ENABLE
def_bool y
depends on X86_32 && !NUMA

so FLATMEM is ruled out via dependency maze.  Be explicit and disable
FLATMEM for DEFERRED_STRUCT_PAGE_INIT so that we do not reintroduce
subtle initialization bugs

[1] http://lkml.kernel.org/r/20160523073157.GD2278@dhcp22.suse.cz

Link: http://lkml.kernel.org/r/1464027356-32282-1-git-send-email-yang.shi@linaro.org
Signed-off-by: Yang Shi <yang.shi@linaro.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/Kconfig

index 2664c118b5d2eea6eee0aae18b9f62c333790a4e..22fa8189e4fc8bce3073ca4db3421aaadb5428e5 100644 (file)
@@ -649,6 +649,7 @@ config DEFERRED_STRUCT_PAGE_INIT
        default n
        depends on ARCH_SUPPORTS_DEFERRED_STRUCT_PAGE_INIT
        depends on MEMORY_HOTPLUG
+       depends on !FLATMEM
        help
          Ordinarily all struct pages are initialised during early boot in a
          single thread. On very large machines this can take a considerable