Re: [PATCH v5 1/4] deactivate invalidated pages

[Minchan Kim]

On Fri, Feb 18, 2011 at 12:50 AM, KAMEZAWA Hiroyuki
<kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> On Fri, 18 Feb 2011 00:08:19 +0900
> Minchan Kim <minchan.kim@xxxxxxxxx> wrote:
>
>> Recently, there are reported problem about thrashing.
>> (http://marc.info/?l=rsync&m=128885034930933&w=2)
>> It happens by backup workloads(ex, nightly rsync).
>> That's because the workload makes just use-once pages
>> and touches pages twice. It promotes the page into
>> active list so that it results in working set page eviction.
>>
>> Some app developer want to support POSIX_FADV_NOREUSE.
>> But other OSes don't support it, either.
>> (http://marc.info/?l=linux-mm&m=128928979512086&w=2)
>>
>> By other approach, app developers use POSIX_FADV_DONTNEED.
>> But it has a problem. If kernel meets page is writing
>> during invalidate_mapping_pages, it can't work.
>> It makes for application programmer to use it since they always
>> have to sync data before calling fadivse(..POSIX_FADV_DONTNEED) to
>> make sure the pages could be discardable. At last, they can't use
>> deferred write of kernel so that they could see performance loss.
>> (http://insights.oetiker.ch/linux/fadvise.html)
>>
>> In fact, invalidation is very big hint to reclaimer.
>> It means we don't use the page any more. So let's move
>> the writing page into inactive list's head if we can't truncate
>> it right now.
>>
>> Why I move page to head of lru on this patch, Dirty/Writeback page
>> would be flushed sooner or later. It can prevent writeout of pageout
>> which is less effective than flusher's writeout.
>>
>> Originally, I reused lru_demote of Peter with some change so added
>> his Signed-off-by.
>>
>> Reported-by: Ben Gamari <bgamari.foss@xxxxxxxxx>
>> Signed-off-by: Minchan Kim <minchan.kim@xxxxxxxxx>
>> Signed-off-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
>> Acked-by: Rik van Riel <riel@xxxxxxxxxx>
>> Acked-by: Mel Gorman <mel@xxxxxxxxx>
>> Reviewed-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>
>> Cc: Wu Fengguang <fengguang.wu@xxxxxxxxx>
>> Cc: Johannes Weiner <hannes@xxxxxxxxxxx>
>> Cc: Nick Piggin <npiggin@xxxxxxxxx>
>> Signed-off-by: Minchan Kim <minchan.kim@xxxxxxxxx>
>
>
> Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx>
>
> One question is ....it seems there is no flush() code for percpu pagevec
> in this patch. Is it safe against cpu hot plug ?
>
> And from memory hot unplug point of view, I'm grad if pagevec for this
> is flushed at the same time as when we clear other per-cpu lru pagevecs.
> (And compaction will be affected by the page_count() magic by pagevec
> Âwhich is flushed only when FADVISE is called.)
>
> Could you add add-on patches for flushing and hooks ?

Isn't it enough in my patch? If I miss your point, Could you elaborate please?

 * Drain pages out of the cpu's pagevecs.
 * Either "cpu" is the current CPU, and preemption has already been
 * disabled; or "cpu" is being hot-unplugged, and is already dead.
@@ -372,6 +427,29 @@ static void drain_cpu_pagevecs(int cpu)
               pagevec_move_tail(pvec);
               local_irq_restore(flags);
       }
+
+       pvec = &per_cpu(lru_deactivate_pvecs, cpu);
+       if (pagevec_count(pvec))
+               ____pagevec_lru_deactivate(pvec);
+}


>
> Thanks,
> -Kame
>
>
>
>> ---
>> Changelog since v4:
>> Â- Change function comments - suggested by Johannes
>> Â- Change function name - suggested by Johannes
>> Â- Drop only dirty/writeback pages to deactive pagevec - suggested by Johannes
>> Â- Add acked-by
>>
>> Changelog since v3:
>> Â- Change function comments - suggested by Johannes
>> Â- Change function name - suggested by Johannes
>> Â- add only dirty/writeback pages to deactive pagevec
>>
>> Changelog since v2:
>> Â- mapped page leaves alone - suggested by Mel
>> Â- pass part related PG_reclaim in next patch.
>>
>> Changelog since v1:
>> Â- modify description
>> Â- correct typo
>> Â- add some comment
>>
>> Âinclude/linux/swap.h | Â Â1 +
>> Âmm/swap.c      Â|  78 ++++++++++++++++++++++++++++++++++++++++++++++++++
>> Âmm/truncate.c    Â|  17 ++++++++---
>> Â3 files changed, 91 insertions(+), 5 deletions(-)
>>
>> diff --git a/include/linux/swap.h b/include/linux/swap.h
>> index 4d55932..c335055 100644
>> --- a/include/linux/swap.h
>> +++ b/include/linux/swap.h
>> @@ -215,6 +215,7 @@ extern void mark_page_accessed(struct page *);
>> Âextern void lru_add_drain(void);
>> Âextern int lru_add_drain_all(void);
>> Âextern void rotate_reclaimable_page(struct page *page);
>> +extern void deactivate_page(struct page *page);
>> Âextern void swap_setup(void);
>>
>> Âextern void add_page_to_unevictable_list(struct page *page);
>> diff --git a/mm/swap.c b/mm/swap.c
>> index c02f936..4aea806 100644
>> --- a/mm/swap.c
>> +++ b/mm/swap.c
>> @@ -39,6 +39,7 @@ int page_cluster;
>>
>> Âstatic DEFINE_PER_CPU(struct pagevec[NR_LRU_LISTS], lru_add_pvecs);
>> Âstatic DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs);
>> +static DEFINE_PER_CPU(struct pagevec, lru_deactivate_pvecs);
>>
>> Â/*
>> Â * This path almost never happens for VM activity - pages are normally
>> @@ -347,6 +348,60 @@ void add_page_to_unevictable_list(struct page *page)
>> Â}
>>
>> Â/*
>> + * If the page can not be invalidated, it is moved to the
>> + * inactive list to speed up its reclaim. ÂIt is moved to the
>> + * head of the list, rather than the tail, to give the flusher
>> + * threads some time to write it out, as this is much more
>> + * effective than the single-page writeout from reclaim.
>> + */
>> +static void lru_deactivate(struct page *page, struct zone *zone)
>> +{
>> + Â Â int lru, file;
>> +
>> + Â Â if (!PageLRU(page) || !PageActive(page))
>> + Â Â Â Â Â Â return;
>> +
>> + Â Â /* Some processes are using the page */
>> + Â Â if (page_mapped(page))
>> + Â Â Â Â Â Â return;
>> +
>> + Â Â file = page_is_file_cache(page);
>> + Â Â lru = page_lru_base_type(page);
>> + Â Â del_page_from_lru_list(zone, page, lru + LRU_ACTIVE);
>> + Â Â ClearPageActive(page);
>> + Â Â ClearPageReferenced(page);
>> + Â Â add_page_to_lru_list(zone, page, lru);
>> + Â Â __count_vm_event(PGDEACTIVATE);
>> +
>> + Â Â update_page_reclaim_stat(zone, page, file, 0);
>> +}
>> +
>> +static void ____pagevec_lru_deactivate(struct pagevec *pvec)
>> +{
>> + Â Â int i;
>> + Â Â struct zone *zone = NULL;
>> +
>> + Â Â for (i = 0; i < pagevec_count(pvec); i++) {
>> + Â Â Â Â Â Â struct page *page = pvec->pages[i];
>> + Â Â Â Â Â Â struct zone *pagezone = page_zone(page);
>> +
>> + Â Â Â Â Â Â if (pagezone != zone) {
>> + Â Â Â Â Â Â Â Â Â Â if (zone)
>> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â spin_unlock_irq(&zone->lru_lock);
>> + Â Â Â Â Â Â Â Â Â Â zone = pagezone;
>> + Â Â Â Â Â Â Â Â Â Â spin_lock_irq(&zone->lru_lock);
>> + Â Â Â Â Â Â }
>> + Â Â Â Â Â Â lru_deactivate(page, zone);
>> + Â Â }
>> + Â Â if (zone)
>> + Â Â Â Â Â Â spin_unlock_irq(&zone->lru_lock);
>> +
>> + Â Â release_pages(pvec->pages, pvec->nr, pvec->cold);
>> + Â Â pagevec_reinit(pvec);
>> +}
>> +
>> +
>> +/*
>> Â * Drain pages out of the cpu's pagevecs.
>> Â * Either "cpu" is the current CPU, and preemption has already been
>> Â * disabled; or "cpu" is being hot-unplugged, and is already dead.
>> @@ -372,6 +427,29 @@ static void drain_cpu_pagevecs(int cpu)
>> Â Â Â Â Â Â Â pagevec_move_tail(pvec);
>> Â Â Â Â Â Â Â local_irq_restore(flags);
>> Â Â Â }
>> +
>> + Â Â pvec = &per_cpu(lru_deactivate_pvecs, cpu);
>> + Â Â if (pagevec_count(pvec))
>> + Â Â Â Â Â Â ____pagevec_lru_deactivate(pvec);
>> +}
>> +
>> +/**
>> + * deactivate_page - forcefully deactivate a page
>> + * @page: page to deactivate
>> + *
>> + * This function hints the VM that @page is a good reclaim candidate,
>> + * for example if its invalidation fails due to the page being dirty
>> + * or under writeback.
>> + */
>> +void deactivate_page(struct page *page)
>> +{
>> + Â Â if (likely(get_page_unless_zero(page))) {
>> + Â Â Â Â Â Â struct pagevec *pvec = &get_cpu_var(lru_deactivate_pvecs);
>> +
>> + Â Â Â Â Â Â if (!pagevec_add(pvec, page))
>> + Â Â Â Â Â Â Â Â Â Â ____pagevec_lru_deactivate(pvec);
>> + Â Â Â Â Â Â put_cpu_var(lru_deactivate_pvecs);
>> + Â Â }
>> Â}
>>
>> Âvoid lru_add_drain(void)
>> diff --git a/mm/truncate.c b/mm/truncate.c
>> index 4d415b3..9ec7bc5 100644
>> --- a/mm/truncate.c
>> +++ b/mm/truncate.c
>> @@ -328,11 +328,12 @@ EXPORT_SYMBOL(truncate_inode_pages);
>> Â * pagetables.
>> Â */
>> Âunsigned long invalidate_mapping_pages(struct address_space *mapping,
>> - Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â Âpgoff_t start, pgoff_t end)
>> + Â Â Â Â Â Â pgoff_t start, pgoff_t end)
>> Â{
>> Â Â Â struct pagevec pvec;
>> Â Â Â pgoff_t next = start;
>> - Â Â unsigned long ret = 0;
>> + Â Â unsigned long ret;
>> + Â Â unsigned long count = 0;
>> Â Â Â int i;
>>
>> Â Â Â pagevec_init(&pvec, 0);
>> @@ -359,8 +360,14 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
>> Â Â Â Â Â Â Â Â Â Â Â if (lock_failed)
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â continue;
>>
>> - Â Â Â Â Â Â Â Â Â Â ret += invalidate_inode_page(page);
>> -
>> + Â Â Â Â Â Â Â Â Â Â ret = invalidate_inode_page(page);
>> + Â Â Â Â Â Â Â Â Â Â /*
>> + Â Â Â Â Â Â Â Â Â Â Â* Invalidation is a hint that the page is no longer
>> + Â Â Â Â Â Â Â Â Â Â Â* of interest and try to speed up its reclaim.
>> + Â Â Â Â Â Â Â Â Â Â Â*/
>> + Â Â Â Â Â Â Â Â Â Â if (!ret)
>> + Â Â Â Â Â Â Â Â Â Â Â Â Â Â deactivate_page(page);
>> + Â Â Â Â Â Â Â Â Â Â count += ret;
>> Â Â Â Â Â Â Â Â Â Â Â unlock_page(page);
>> Â Â Â Â Â Â Â Â Â Â Â if (next > end)
>> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â break;
>> @@ -369,7 +376,7 @@ unsigned long invalidate_mapping_pages(struct address_space *mapping,
>> Â Â Â Â Â Â Â mem_cgroup_uncharge_end();
>> Â Â Â Â Â Â Â cond_resched();
>> Â Â Â }
>> - Â Â return ret;
>> + Â Â return count;
>> Â}
>> ÂEXPORT_SYMBOL(invalidate_mapping_pages);
>>
>> --
>> 1.7.1
>>
>>
>
>



-- 
Kind regards,
Minchan Kim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/