Re: [RFC][PATCH 1/2] memcg: avoiding unnecessary get_page at move_charge

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



2010/10/16 Christoph Lameter <cl@xxxxxxxxx>:
> On Fri, 15 Oct 2010, KAMEZAWA Hiroyuki wrote:
>
>> But above is called all under pte_offset_map_lock().
>> get_page_unless_zero() #1 is not necessary because we do all under a
>> pte_offset_map_lock().
>
> The two (ptl and refcount) are entirely different. The ptl is for
> protecting the page table. The refcount handles only the page.
>
> However, if the entry in the page table is pointing to the page then there
> must have been a refcount taken on the page. So if you know that the page
> is in the page table and you took the ptl then you can be sure that the
> page refcount will not become zero. Therefore get_page_unless_zero() will
> never fail and there is no need to take additional refcounts as long as
> the page table lock is held and the page is not removed from the page
> table.
>

Ok, thank you for explanation. I can make this function faster.

>> Index: mmotm-1013/mm/vmscan.c
>> ===================================================================
>> --- mmotm-1013.orig/mm/vmscan.c
>> +++ mmotm-1013/mm/vmscan.c
>> @@ -1166,7 +1166,8 @@ static unsigned long clear_active_flags(
>>   * found will be decremented.
>>   *
>>   * Restrictions:
>> - * (1) Must be called with an elevated refcount on the page. This is a
>> + * (1) Must be called with an elevated refcount on the page, IOW, the
>> + *     caller must guarantee that there is a stable reference. This is a
>>   *     fundamentnal difference from isolate_lru_pages (which is called
>>   *     without a stable reference).
>>   * (2) the lru_lock must not be held.
>
> There is no need for this change since you have an elevated refcount.
> IMH The words "stable reference" may be confusing since the refcount may
> change. The elevated refcount protects against the freeing of the page.
>

Sure, drop change this in v2. I misunderstand "elevated refcount"
means "extra get_page()".

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href



[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux]     [Linux OMAP]     [Linux MIPS]     [ECOS]     [Asterisk Internet PBX]     [Linux API]