From: Changsheng Liu <[email protected]>
After the user config CONFIG_MOVABLE_NODE and movable_node kernel option,
When the memory is hot added, should_add_memory_movable() return 0
because all zones including movable zone are empty,
so the memory that was hot added will be added to the normal zone
and the normal zone will be created firstly.
But we want the whole node to be added to movable zone defaultly.
So we change should_add_memory_movable(): if the user config
CONFIG_MOVABLE_NODE and movable_node kernel option
it will always return 1 and all zones is empty at the same time,
so that the movable zone will be created firstly
and then the whole node will be added to movable zone defaultly.
If we want the node to be added to normal zone,
we can do it as follows:
"echo online_kernel > /sys/devices/system/memory/memoryXXX/state"
If the memory is added to movable zone defaultly,
the user can offline it and add it to other zone again.
But if the memory is added to normal zone defaultly,
the user will not offline the memory used by kernel.
Reviewed-by: Andrew Morton <[email protected]>
Signed-off-by: Changsheng Liu <[email protected]>i
Tested-by: Dongdong Fan <[email protected]>
---
mm/memory_hotplug.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 26fbba7..b5f14fa 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -1198,6 +1198,9 @@ static int should_add_memory_movable(int nid, u64 start, u64 size)
pg_data_t *pgdat = NODE_DATA(nid);
struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
+ if (movable_node_is_enabled())
+ return 1;
+
if (zone_is_empty(movable_zone))
return 0;
--
1.7.1
On 08/27/2015 10:54 AM, Changsheng Liu wrote:
> From: Changsheng Liu <[email protected]>
>
> After the user config CONFIG_MOVABLE_NODE and movable_node kernel option,
> When the memory is hot added, should_add_memory_movable() return 0
> because all zones including movable zone are empty,
> so the memory that was hot added will be added to the normal zone
> and the normal zone will be created firstly.
> But we want the whole node to be added to movable zone defaultly.
OK it seems current behavior indeed goes against the expectations of
setting movable_node.
> So we change should_add_memory_movable(): if the user config
> CONFIG_MOVABLE_NODE and movable_node kernel option
> it will always return 1 and all zones is empty at the same time,
> so that the movable zone will be created firstly
> and then the whole node will be added to movable zone defaultly.
> If we want the node to be added to normal zone,
> we can do it as follows:
> "echo online_kernel > /sys/devices/system/memory/memoryXXX/state"
>
> If the memory is added to movable zone defaultly,
> the user can offline it and add it to other zone again.
Was this tested to really work as well? Per Yasuaki's explanation in v2,
you shouldn't create ZONE_MOVABLE before ZONE_NORMAL.
> But if the memory is added to normal zone defaultly,
> the user will not offline the memory used by kernel.
>
> Reviewed-by: Andrew Morton <[email protected]>
Interesting...
> Signed-off-by: Changsheng Liu <[email protected]>i
> Tested-by: Dongdong Fan <[email protected]>
> ---
> mm/memory_hotplug.c | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 26fbba7..b5f14fa 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -1198,6 +1198,9 @@ static int should_add_memory_movable(int nid, u64 start, u64 size)
> pg_data_t *pgdat = NODE_DATA(nid);
> struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
>
> + if (movable_node_is_enabled())
> + return 1;
> +
> if (zone_is_empty(movable_zone))
> return 0;
>
>
在 2015/8/27 17:29, Vlastimil Babka 写道:
> On 08/27/2015 10:54 AM, Changsheng Liu wrote:
>> From: Changsheng Liu <[email protected]>
>>
>> After the user config CONFIG_MOVABLE_NODE and movable_node kernel
>> option,
>> When the memory is hot added, should_add_memory_movable() return 0
>> because all zones including movable zone are empty,
>> so the memory that was hot added will be added to the normal zone
>> and the normal zone will be created firstly.
>> But we want the whole node to be added to movable zone defaultly.
>
> OK it seems current behavior indeed goes against the expectations of
> setting movable_node.
>
>> So we change should_add_memory_movable(): if the user config
>> CONFIG_MOVABLE_NODE and movable_node kernel option
>> it will always return 1 and all zones is empty at the same time,
>> so that the movable zone will be created firstly
>> and then the whole node will be added to movable zone defaultly.
>> If we want the node to be added to normal zone,
>> we can do it as follows:
>> "echo online_kernel > /sys/devices/system/memory/memoryXXX/state"
>>
>> If the memory is added to movable zone defaultly,
>> the user can offline it and add it to other zone again.
>
> Was this tested to really work as well? Per Yasuaki's explanation in
> v2, you shouldn't create ZONE_MOVABLE before ZONE_NORMAL.
We will test it more fully.
There is on limit written in Documentation/memory-hotplug.txt when
move the memory from movable zone to normal zone:
echo online_kernel > /sys/devices/system/memory/memoryXXX/state
(NOTE: current limit: this memory block must be adjacent to
ZONE_NORMAL)
The zone will be created as follows:
First,all zones have been initialized,but all zones are empty, then
the should_add_memory_movable() return 1 so the movable is created and
add the whole node to movable zone.
empty whole node's memory
|-------------------|-------------------------------|---
ZONE_NORMAL ZONE_MOVABLE
when we move the memory to normal zone
empty whole node's memory
|--------------------|----------------------------------------------------:-------------------------------------------|
one memory block ajacent to
ZONE_NORMAL :
ZONE_NORMAL ZONE_MOVABLE :
:
one memory block
: whole node - one
memory block
|-------------------------------------------------------------------------|-------------------------------------------|
ZONE_NORMAL ZONE_MOVABLE
>
>> But if the memory is added to normal zone defaultly,
>> the user will not offline the memory used by kernel.
>>
>> Reviewed-by: Andrew Morton <[email protected]>
>
> Interesting...
>
>> Signed-off-by: Changsheng Liu <[email protected]>i
>> Tested-by: Dongdong Fan <[email protected]>
>> ---
>> mm/memory_hotplug.c | 3 +++
>> 1 files changed, 3 insertions(+), 0 deletions(-)
>>
>> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
>> index 26fbba7..b5f14fa 100644
>> --- a/mm/memory_hotplug.c
>> +++ b/mm/memory_hotplug.c
>> @@ -1198,6 +1198,9 @@ static int should_add_memory_movable(int nid,
>> u64 start, u64 size)
>> pg_data_t *pgdat = NODE_DATA(nid);
>> struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
>>
>> + if (movable_node_is_enabled())
>> + return 1;
>> +
>> if (zone_is_empty(movable_zone))
>> return 0;
>>
>>
>
> .
>
On Thu, 27 Aug 2015 04:54:01 -0400
Changsheng Liu <[email protected]> wrote:
> From: Changsheng Liu <[email protected]>
>
> After the user config CONFIG_MOVABLE_NODE and movable_node kernel option,
> When the memory is hot added, should_add_memory_movable() return 0
> because all zones including movable zone are empty,
> so the memory that was hot added will be added to the normal zone
> and the normal zone will be created firstly.
> But we want the whole node to be added to movable zone defaultly.
>
> So we change should_add_memory_movable(): if the user config
> CONFIG_MOVABLE_NODE and movable_node kernel option
> it will always return 1 and all zones is empty at the same time,
> so that the movable zone will be created firstly
> and then the whole node will be added to movable zone defaultly.
> If we want the node to be added to normal zone,
> we can do it as follows:
> "echo online_kernel > /sys/devices/system/memory/memoryXXX/state"
>
> If the memory is added to movable zone defaultly,
> the user can offline it and add it to other zone again.
> But if the memory is added to normal zone defaultly,
> the user will not offline the memory used by kernel.
>
> Reviewed-by: Andrew Morton <[email protected]>
> Signed-off-by: Changsheng Liu <[email protected]>i
> Tested-by: Dongdong Fan <[email protected]>
> ---
> mm/memory_hotplug.c | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 26fbba7..b5f14fa 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
> @@ -1198,6 +1198,9 @@ static int should_add_memory_movable(int nid, u64 start, u64 size)
> pg_data_t *pgdat = NODE_DATA(nid);
> struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
>
> + if (movable_node_is_enabled())
> + return 1;
> +
> if (zone_is_empty(movable_zone))
> return 0;
Your patch cannot prevent from creating ZONE_MOVABLE before/in
ZONE_NORAML. Please check that start PFN of hot added memory is
after ZONE_NORAML.
Thanks,
Yasuaki Ishimatsu
>
> --
> 1.7.1
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/