On 2015/6/30 15:53, Kamezawa Hiroyuki wrote:
> On 2015/06/30 11:45, Xishi Qiu wrote:
>> On 2015/6/29 15:32, Kamezawa Hiroyuki wrote:
>>
>>> On 2015/06/27 11:24, Xishi Qiu wrote:
>>>> This patch introduces a new migratetype called "MIGRATE_MIRROR", it is used to
>>>> allocate mirrored pages.
>>>> When cat /proc/pagetypeinfo, you can see the count of free mirrored blocks.
>>>>
>>>> Signed-off-by: Xishi Qiu <qiuxishi@xxxxxxxxxx>
>>>
>>> My fear about this approarch is that this may break something existing.
>>>
>>> Now, when we add MIGRATE_MIRROR type, we'll hide attributes of pageblocks as
>>> MIGRATE_UNMOVABOLE, MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE.
>>>
>>> Logically, MIRROR attribute is independent from page mobility and this overwrites
>>> will make some information lost.
>>>
>>> Then,
>>>
>>>> ---
>>>> include/linux/mmzone.h | 9 +++++++++
>>>> mm/page_alloc.c | 3 +++
>>>> mm/vmstat.c | 3 +++
>>>> 3 files changed, 15 insertions(+)
>>>>
>>>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>>>> index 54d74f6..54e891a 100644
>>>> --- a/include/linux/mmzone.h
>>>> +++ b/include/linux/mmzone.h
>>>> @@ -39,6 +39,9 @@ enum {
>>>> MIGRATE_UNMOVABLE,
>>>> MIGRATE_RECLAIMABLE,
>>>> MIGRATE_MOVABLE,
>>>> +#ifdef CONFIG_MEMORY_MIRROR
>>>> + MIGRATE_MIRROR,
>>>> +#endif
>>>
>>> I think
>>> MIGRATE_MIRROR_UNMOVABLE,
>>> MIGRATE_MIRROR_RECLAIMABLE,
>>> MIGRATE_MIRROR_MOVABLE, <== adding this may need discuss.
>>> MIGRATE_MIRROR_RESERVED, <== reserved pages should be maintained per mirrored/unmirrored.
>>>
>>
>> Hi Kame,
>>
>> You mean add 3 or 4 new migratetype?
>>
>
> yes. But please check how NR_MIGRATETYPE_BITS will be.
> I think this will not have big impact in x86-64 .
>
>>> should be added with the following fallback list.
>>>
>>> /*
>>> * MIRROR page range is defined by firmware at boot. The range is limited
>>> * and is used only for kernel memory mirroring.
>>> */
>>> [MIGRATE_UNMOVABLE_MIRROR] = {MIGRATE_RECLAIMABLE_MIRROR, MIGRATE_RESERVE}
>>> [MIGRATE_RECLAIMABLE_MIRROR] = {MIGRATE_UNMOVABLE_MIRROR, MIGRATE_RESERVE}
>>>
>>
>> Why not like this:
>> {MIGRATE_RECLAIMABLE_MIRROR, MIGRATE_MIRROR_RESERVED, MIGRATE_RESERVE}
>>
>
> My mistake.
> [MIGRATE_UNMOVABLE_MIRROR] = {MIGRATE_RECLAIMABLE_MIRROR, MIGRATE_RESERVE_MIRROR}
> [MIGRATE_RECLAIMABLE_MIRROR] = {MIGRATE_UNMOVABLE_MIRROR, MIGRATE_RESERVE_MIRROR}
>
> was my intention. This means mirrored memory and unmirrored memory is separated completely.
>
> But this should affect kswapd or other memory reclaim logic.
>
> for example, kswapd stops free pages are more than hi watermark.
> But mirrored/unmirrored pages exhausted cases are not handled in this series.
> You need some extra check in memory reclaim logic if you go with migration_type.
>
OK, I understand. Thank you for your suggestion.
Thanks,
Xishi Qiu
>
>
>>> Then, we'll not lose the original information of "Reclaiable Pages".
>>>
>>> One problem here is whteher we should have MIGRATE_RESERVE_MIRROR.
>>>
>>> If we never allow users to allocate mirrored memory, we should have MIGRATE_RESERVE_MIRROR.
>>> But it seems to require much more code change to do that.
>>>
>>> Creating a zone or adding an attribues to zones are another design choice.
>>>
>>
>> If we add a new zone, mirror_zone will span others, I'm worry about this
>> maybe have problems.
>
> Yes. that's problem. And zoneid bit is very limited resource.
> (....But memory reclaim logic can be unchanged.)
>
> Anyway, I'd like to see your solution with above changes 1st rather than adding zones.
>
> Thanks,
> -Kame
>
>
>
> .
>
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>