Re: [v2 PATCH] mm: move page_ext_init after all struct pages are initialized

[Shi, Yang]

On 5/23/2016 12:31 AM, Michal Hocko wrote:

On Fri 20-05-16 08:41:09, Shi, Yang wrote:

On 5/20/2016 6:16 AM, Michal Hocko wrote:

On Thu 19-05-16 15:13:26, Yang Shi wrote:
[...]

diff --git a/init/main.c b/init/main.c
index b3c6e36..2075faf 100644
--- a/init/main.c
+++ b/init/main.c
@@ -606,7 +606,6 @@ asmlinkage __visible void __init start_kernel(void)
                initrd_start = 0;
        }
 #endif
-       page_ext_init();
        debug_objects_mem_init();
        kmemleak_init();
        setup_per_cpu_pageset();
@@ -1004,6 +1003,8 @@ static noinline void __init kernel_init_freeable(void)
        sched_init_smp();

        page_alloc_init_late();
+       /* Initialize page ext after all struct pages are initializaed */
+       page_ext_init();

        do_basic_setup();

I might be missing something but don't we have the same problem with
CONFIG_FLATMEM? page_ext_init_flatmem is called way earlier. Or
CONFIG_DEFERRED_STRUCT_PAGE_INIT is never enabled for CONFIG_FLATMEM?

Yes, CONFIG_DEFERRED_STRUCT_PAGE_INIT depends on MEMORY_HOTPLUG which
depends on SPARSEMEM. So, this config is not valid for FLATMEM at all.

Well
config MEMORY_HOTPLUG
        bool "Allow for memory hot-add"
        depends on SPARSEMEM || X86_64_ACPI_NUMA
        depends on ARCH_ENABLE_MEMORY_HOTPLUG

I wasn't really sure about X86_64_ACPI_NUMA dependency branch which
depends on X86_64 && NUMA && ACPI && PCI and that didn't sound like
SPARSEMEM only. If the FLATMEM shouldn't exist with

Actually, FLATMEMT depends on !NUMA.

CONFIG_DEFERRED_STRUCT_PAGE_INIT can we make that explicit please?

Sure, it makes the condition clearer and more readable.

Thanks,
Yang