2013-04-02 10:12:48

by Lin Feng

Subject: [PATCH] x86: numa: mm: kill double initialization for NODE_DATA

We initialize node_id, node_start_pfn and node_spanned_pages for NODE_DATA in
initmem_init() while the later two members are kept unused and will be
recaculated soon in paging_init(), so remove the useless assignments.

PS. For clarifying calling chains are showed as follows:
setup_arch()
...
initmem_init()
x86_numa_init()
numa_init()
numa_register_memblks()
setup_node_data()
NODE_DATA(nid)->node_id = nid;
NODE_DATA(nid)->node_start_pfn = start >> PAGE_SHIFT;
NODE_DATA(nid)->node_spanned_pages = (end - start) >> PAGE_SHIFT;
...
x86_init.paging.pagetable_init()
paging_init()
...
sparse_init()
sparse_early_usemaps_alloc_node()
sparse_early_usemaps_alloc_pgdat_section()
___alloc_bootmem_node_nopanic()
__alloc_memory_core_early(pgdat->node_id,...)
...
zone_sizes_init()
free_area_init_nodes()
free_area_init_node()
pgdat->node_id = nid;
pgdat->node_start_pfn = node_start_pfn;
calculate_node_totalpages();
pgdat->node_spanned_pages = totalpages;


Signed-off-by: Lin Feng <[email protected]>
---
arch/x86/mm/numa.c | 2 --
1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/numa.c b/arch/x86/mm/numa.c
index 72fe01e..efdd08f 100644
--- a/arch/x86/mm/numa.c
+++ b/arch/x86/mm/numa.c
@@ -230,8 +230,6 @@ static void __init setup_node_data(int nid, u64 start, u64 end)
node_data[nid] = nd;
memset(NODE_DATA(nid), 0, sizeof(pg_data_t));
NODE_DATA(nid)->node_id = nid;
- NODE_DATA(nid)->node_start_pfn = start >> PAGE_SHIFT;
- NODE_DATA(nid)->node_spanned_pages = (end - start) >> PAGE_SHIFT;

node_set_online(nid);
}
--
1.7.1

2013-04-03 01:52:39

by Lin Feng

Subject: Re: [PATCH] x86: numa: mm: kill double initialization for NODE_DATA

Hi Wanpeng,

On 04/02/2013 06:57 PM, Wanpeng Li wrote:
>> >PS. For clarifying calling chains are showed as follows:
>> >setup_arch()
>> > ...
>> > initmem_init()
>> > x86_numa_init()
>> > numa_init()
>> > numa_register_memblks()
>> > setup_node_data()
>> > NODE_DATA(nid)->node_id = nid;
>> > NODE_DATA(nid)->node_start_pfn = start >> PAGE_SHIFT;
>> > NODE_DATA(nid)->node_spanned_pages = (end - start) >> PAGE_SHIFT;
>> > ...
>> > x86_init.paging.pagetable_init()
>> > paging_init()
>> > ...
>> > sparse_init()
>> > sparse_early_usemaps_alloc_node()
>> > sparse_early_usemaps_alloc_pgdat_section()
>> > ___alloc_bootmem_node_nopanic()
>> > __alloc_memory_core_early(pgdat->node_id,...)
>> > ...
>> > zone_sizes_init()
>> > free_area_init_nodes()
>> > free_area_init_node()
>> > pgdat->node_id = nid;
>> > pgdat->node_start_pfn = node_start_pfn;
>> > calculate_node_totalpages();
>> > pgdat->node_spanned_pages = totalpages;
>> >
> You miss the nodes which could become online at some point, but not
> online currently.

Sorry, I'm not quite understanding what you said.

I keep node_set_online(nid) there. In boot phase if a node is online now it wil be
reinitialized later by zone_sizes_init() else if a node is hotpluged after system is
up it will also be initialized by hotadd_new_pgdat() which falls into calling
free_area_init_node().

Besides this I'm not sure there are any other dependency besides what you worry about,
while I tested this on a x86_64 numa system with hot-add nodes and the meminfo statics
looks right before and after hot-add memory.

thanks for your patient,
linfeng