Re: [PATCH] kswapd: no need reclaim cma pages triggered by unmovable allocation

[zhou xianrong <zhouxianrong@tom.com>]

在 2021/3/15 23:46, David Hildenbrand 写道:
> On 13.03.21 09:31, zhou wrote:
>> From: zhou xianrong <xianrong.zhou@transsion.com>
>>
>> For purpose of better migration cma pages are allocated after
>> failure movalbe allocations and are used normally for file pages
>> or anonymous pages.
>
> I failed to parse that senctence.
>
> "For better migration, CMA pages are allocated after failing 
> allocation of movable allocations and are used for backing files or 
> anonymous memory."
>
> Still doesn't make any sense to me. Can you clarify?
>
I mean alloc_page() allocates cma pages using fallback style so that 
alloc_contig_range() can allocate it easily.
>
>>
>> In reclaim path many cma pages if configurated are reclaimed
>
> s/configurated/configured/
>
>> from lru lists in kswapd mainly or direct reclaim triggered by
>> unmovable or reclaimable allocations. But these reclaimed cma
>> pages can not be used by original unmovable or reclaimable
>> allocations. So the reclaim are unnecessary.
>
> Might be a dump question, but why can't reclaimable allocations end up 
> on CMA? (for unmovable allocations, this is clear) Or did I 
> misunderstand what that paragraph was trying to tell me?
>
I remember only movable migratetype can fallback into cma.
>>
>> So the unmovable or reclaimable allocations should not trigger
>> reclaiming cma pages. The patch adds third factor of migratetype
>> which is just like factors of zone index or order kswapd need
>> consider. The modification follows codes of zone index
>> consideration. And it is straightforward that skips reclaiming
>> cma pages in reclaim procedure which is triggered only by
>> unmovable or reclaimable allocations.
>>
>> This optimization can avoid ~3% unnecessary isolations from cma
>> (cma isolated / total isolated) with configuration of total 100Mb
>> cma pages.
>
> Can you say a few words about interaction with ZONE_MOVABLE, which 
> behaves similar to CMA? I.e., does the same apply to ZONE_MOVABLE? Is 
> it already handled?
>
>>
>> Signed-off-by: zhou xianrong <xianrong.zhou@transsion.com>
>> Signed-off-by: feng ruxian <ruxian.feng@transsion.com>
>> ---
>>   include/linux/mmzone.h        |  6 ++--
>>   include/trace/events/vmscan.h | 20 +++++++----
>>   mm/page_alloc.c               |  5 +--
>>   mm/vmscan.c                   | 63 +++++++++++++++++++++++++++++------
>>   4 files changed, 73 insertions(+), 21 deletions(-)
>>
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index b593316bff3d..7dd38d7372b9 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -301,6 +301,8 @@ struct lruvec {
>>   #define ISOLATE_ASYNC_MIGRATE    ((__force isolate_mode_t)0x4)
>>   /* Isolate unevictable pages */
>>   #define ISOLATE_UNEVICTABLE    ((__force isolate_mode_t)0x8)
>> +/* Isolate none cma pages */
>> +#define ISOLATE_NONCMA        ((__force isolate_mode_t)0x10)
>>     /* LRU Isolation modes. */
>>   typedef unsigned __bitwise isolate_mode_t;
>> @@ -756,7 +758,7 @@ typedef struct pglist_data {
>>       wait_queue_head_t pfmemalloc_wait;
>>       struct task_struct *kswapd;    /* Protected by
>>                          mem_hotplug_begin/end() */
>> -    int kswapd_order;
>> +    int kswapd_order, kswapd_migratetype;
>>       enum zone_type kswapd_highest_zoneidx;
>>         int kswapd_failures;        /* Number of 'reclaimed == 0' 
>> runs */
>> @@ -840,7 +842,7 @@ static inline bool pgdat_is_empty(pg_data_t *pgdat)
>>     void build_all_zonelists(pg_data_t *pgdat);
>>   void wakeup_kswapd(struct zone *zone, gfp_t gfp_mask, int order,
>> -           enum zone_type highest_zoneidx);
>> +           int migratetype, enum zone_type highest_zoneidx);
>>   bool __zone_watermark_ok(struct zone *z, unsigned int order, 
>> unsigned long mark,
>>                int highest_zoneidx, unsigned int alloc_flags,
>>                long free_pages);
>> diff --git a/include/trace/events/vmscan.h 
>> b/include/trace/events/vmscan.h
>> index 2070df64958e..41bbafdfde84 100644
>> --- a/include/trace/events/vmscan.h
>> +++ b/include/trace/events/vmscan.h
>> @@ -51,37 +51,41 @@ TRACE_EVENT(mm_vmscan_kswapd_sleep,
>>     TRACE_EVENT(mm_vmscan_kswapd_wake,
>>   -    TP_PROTO(int nid, int zid, int order),
>> +    TP_PROTO(int nid, int zid, int order, int mt),
>>   -    TP_ARGS(nid, zid, order),
>> +    TP_ARGS(nid, zid, order, mt),
>>         TP_STRUCT__entry(
>>           __field(    int,    nid    )
>>           __field(    int,    zid    )
>>           __field(    int,    order    )
>> +        __field(    int,    mt    )
>>       ),
>>         TP_fast_assign(
>>           __entry->nid    = nid;
>>           __entry->zid    = zid;
>>           __entry->order    = order;
>> +        __entry->mt    = mt;
>>       ),
>>   -    TP_printk("nid=%d order=%d",
>> +    TP_printk("nid=%d order=%d migratetype=%d",
>>           __entry->nid,
>> -        __entry->order)
>> +        __entry->order,
>> +        __entry->mt)
>>   );
>>     TRACE_EVENT(mm_vmscan_wakeup_kswapd,
>>   -    TP_PROTO(int nid, int zid, int order, gfp_t gfp_flags),
>> +    TP_PROTO(int nid, int zid, int order, int mt, gfp_t gfp_flags),
>>   -    TP_ARGS(nid, zid, order, gfp_flags),
>> +    TP_ARGS(nid, zid, order, mt, gfp_flags),
>>         TP_STRUCT__entry(
>>           __field(    int,    nid        )
>>           __field(    int,    zid        )
>>           __field(    int,    order        )
>> +        __field(    int,    mt        )
>>           __field(    gfp_t,    gfp_flags    )
>>       ),
>>   @@ -89,12 +93,14 @@ TRACE_EVENT(mm_vmscan_wakeup_kswapd,
>>           __entry->nid        = nid;
>>           __entry->zid        = zid;
>>           __entry->order        = order;
>> +        __entry->mt        = mt;
>>           __entry->gfp_flags    = gfp_flags;
>>       ),
>>   -    TP_printk("nid=%d order=%d gfp_flags=%s",
>> +    TP_printk("nid=%d order=%d migratetype=%d gfp_flags=%s",
>>           __entry->nid,
>>           __entry->order,
>> +        __entry->mt,
>>           show_gfp_flags(__entry->gfp_flags))
>>   );
>>   diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 519a60d5b6f7..45ceb15721b8 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -3517,7 +3517,7 @@ struct page *rmqueue(struct zone *preferred_zone,
>>       /* Separate test+clear to avoid unnecessary atomics */
>>       if (test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags)) {
>>           clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags);
>> -        wakeup_kswapd(zone, 0, 0, zone_idx(zone));
>> +        wakeup_kswapd(zone, 0, 0, migratetype, zone_idx(zone));
>>       }
>>         VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
>> @@ -4426,11 +4426,12 @@ static void wake_all_kswapds(unsigned int 
>> order, gfp_t gfp_mask,
>>       struct zone *zone;
>>       pg_data_t *last_pgdat = NULL;
>>       enum zone_type highest_zoneidx = ac->highest_zoneidx;
>> +    int migratetype = ac->migratetype;
>>         for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, 
>> highest_zoneidx,
>>                       ac->nodemask) {
>>           if (last_pgdat != zone->zone_pgdat)
>> -            wakeup_kswapd(zone, gfp_mask, order, highest_zoneidx);
>> +            wakeup_kswapd(zone, gfp_mask, order, migratetype, 
>> highest_zoneidx);
>>           last_pgdat = zone->zone_pgdat;
>>       }
>>   }
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index b1b574ad199d..184f0c4c7151 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -99,6 +99,9 @@ struct scan_control {
>>       /* Can pages be swapped as part of reclaim? */
>>       unsigned int may_swap:1;
>>   +    /* Can cma pages be reclaimed? */
>> +    unsigned int may_cma:1;
>> +
>>       /*
>>        * Cgroups are not reclaimed below their configured memory.low,
>>        * unless we threaten to OOM. If any cgroups are skipped due to
>> @@ -286,6 +289,11 @@ static bool writeback_throttling_sane(struct 
>> scan_control *sc)
>>   }
>>   #endif
>>   +static bool movable_reclaim(gfp_t gfp_mask)
>> +{
>> +    return is_migrate_movable(gfp_migratetype(gfp_mask));
>> +}
>> +
>>   /*
>>    * This misses isolated pages which are not accounted for to save 
>> counters.
>>    * As the data only determines if reclaim or compaction continues, 
>> it is
>> @@ -1499,6 +1507,7 @@ unsigned int 
>> reclaim_clean_pages_from_list(struct zone *zone,
>>           .gfp_mask = GFP_KERNEL,
>>           .priority = DEF_PRIORITY,
>>           .may_unmap = 1,
>> +        .may_cma = 1,
>>       };
>>       struct reclaim_stat stat;
>>       unsigned int nr_reclaimed;
>> @@ -1593,6 +1602,9 @@ int __isolate_lru_page_prepare(struct page 
>> *page, isolate_mode_t mode)
>>       if ((mode & ISOLATE_UNMAPPED) && page_mapped(page))
>>           return ret;
>>   +    if ((mode & ISOLATE_NONCMA) && 
>> is_migrate_cma(get_pageblock_migratetype(page)))
>> +        return ret;
>> +
>>       return 0;
>>   }
>>   @@ -1647,7 +1659,10 @@ static unsigned long 
>> isolate_lru_pages(unsigned long nr_to_scan,
>>       unsigned long skipped = 0;
>>       unsigned long scan, total_scan, nr_pages;
>>       LIST_HEAD(pages_skipped);
>> -    isolate_mode_t mode = (sc->may_unmap ? 0 : ISOLATE_UNMAPPED);
>> +    isolate_mode_t mode;
>> +
>> +    mode = (sc->may_unmap ? 0 : ISOLATE_UNMAPPED);
>> +    mode |= (sc->may_cma ? 0 : ISOLATE_NONCMA);
>>         total_scan = 0;
>>       scan = 0;
>> @@ -2125,6 +2140,7 @@ unsigned long reclaim_pages(struct list_head 
>> *page_list)
>>           .may_writepage = 1,
>>           .may_unmap = 1,
>>           .may_swap = 1,
>> +        .may_cma = 1,
>>       };
>>         while (!list_empty(page_list)) {
>> @@ -3253,6 +3269,7 @@ unsigned long try_to_free_pages(struct zonelist 
>> *zonelist, int order,
>>           .may_writepage = !laptop_mode,
>>           .may_unmap = 1,
>>           .may_swap = 1,
>> +        .may_cma = movable_reclaim(gfp_mask),
>>       };
>>         /*
>> @@ -3298,6 +3315,7 @@ unsigned long mem_cgroup_shrink_node(struct 
>> mem_cgroup *memcg,
>>           .may_unmap = 1,
>>           .reclaim_idx = MAX_NR_ZONES - 1,
>>           .may_swap = !noswap,
>> +        .may_cma = 1,
>>       };
>>         WARN_ON_ONCE(!current->reclaim_state);
>> @@ -3341,6 +3359,7 @@ unsigned long 
>> try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg,
>>           .may_writepage = !laptop_mode,
>>           .may_unmap = 1,
>>           .may_swap = may_swap,
>> +        .may_cma = 1,
>>       };
>>       /*
>>        * Traverse the ZONELIST_FALLBACK zonelist of the current node 
>> to put
>> @@ -3548,7 +3567,7 @@ static bool kswapd_shrink_node(pg_data_t *pgdat,
>>    * or lower is eligible for reclaim until at least one usable zone is
>>    * balanced.
>>    */
>> -static int balance_pgdat(pg_data_t *pgdat, int order, int 
>> highest_zoneidx)
>> +static int balance_pgdat(pg_data_t *pgdat, int order, int 
>> migratetype, int highest_zoneidx)
>>   {
>>       int i;
>>       unsigned long nr_soft_reclaimed;
>> @@ -3650,6 +3669,7 @@ static int balance_pgdat(pg_data_t *pgdat, int 
>> order, int highest_zoneidx)
>>            */
>>           sc.may_writepage = !laptop_mode && !nr_boost_reclaim;
>>           sc.may_swap = !nr_boost_reclaim;
>> +        sc.may_cma = is_migrate_movable(migratetype);
>>             /*
>>            * Do some background aging of the anon list, to give
>> @@ -3771,8 +3791,15 @@ static enum zone_type 
>> kswapd_highest_zoneidx(pg_data_t *pgdat,
>>       return curr_idx == MAX_NR_ZONES ? prev_highest_zoneidx : curr_idx;
>>   }
>>   +static int kswapd_migratetype(pg_data_t *pgdat, int prev_migratetype)
>> +{
>> +    int curr_migratetype = READ_ONCE(pgdat->kswapd_migratetype);
>> +
>> +    return curr_migratetype == MIGRATE_TYPES ? prev_migratetype : 
>> curr_migratetype;
>> +}
>> +
>>   static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, 
>> int reclaim_order,
>> -                unsigned int highest_zoneidx)
>> +                int migratetype, unsigned int highest_zoneidx)
>>   {
>>       long remaining = 0;
>>       DEFINE_WAIT(wait);
>> @@ -3807,8 +3834,8 @@ static void kswapd_try_to_sleep(pg_data_t 
>> *pgdat, int alloc_order, int reclaim_o
>>           remaining = schedule_timeout(HZ/10);
>>             /*
>> -         * If woken prematurely then reset kswapd_highest_zoneidx and
>> -         * order. The values will either be from a wakeup request or
>> +         * If woken prematurely then reset kswapd_highest_zoneidx, 
>> order
>> +         * and migratetype. The values will either be from a wakeup 
>> request or
>>            * the previous request that slept prematurely.
>>            */
>>           if (remaining) {
>> @@ -3818,6 +3845,10 @@ static void kswapd_try_to_sleep(pg_data_t 
>> *pgdat, int alloc_order, int reclaim_o
>>                 if (READ_ONCE(pgdat->kswapd_order) < reclaim_order)
>>                   WRITE_ONCE(pgdat->kswapd_order, reclaim_order);
>> +
>> +            if 
>> (!is_migrate_movable(READ_ONCE(pgdat->kswapd_migratetype)))
>> +                WRITE_ONCE(pgdat->kswapd_migratetype,
>> +                        kswapd_migratetype(pgdat, migratetype));
>>           }
>>             finish_wait(&pgdat->kswapd_wait, &wait);
>> @@ -3870,6 +3901,7 @@ static void kswapd_try_to_sleep(pg_data_t 
>> *pgdat, int alloc_order, int reclaim_o
>>    */
>>   static int kswapd(void *p)
>>   {
>> +    int migratetype = 0;
>>       unsigned int alloc_order, reclaim_order;
>>       unsigned int highest_zoneidx = MAX_NR_ZONES - 1;
>>       pg_data_t *pgdat = (pg_data_t*)p;
>> @@ -3895,23 +3927,27 @@ static int kswapd(void *p)
>>       set_freezable();
>>         WRITE_ONCE(pgdat->kswapd_order, 0);
>> +    WRITE_ONCE(pgdat->kswapd_migratetype, MIGRATE_TYPES);
>>       WRITE_ONCE(pgdat->kswapd_highest_zoneidx, MAX_NR_ZONES);
>>       for ( ; ; ) {
>>           bool ret;
>>             alloc_order = reclaim_order = 
>> READ_ONCE(pgdat->kswapd_order);
>> +        migratetype = kswapd_migratetype(pgdat, migratetype);
>>           highest_zoneidx = kswapd_highest_zoneidx(pgdat,
>>                               highest_zoneidx);
>>     kswapd_try_sleep:
>>           kswapd_try_to_sleep(pgdat, alloc_order, reclaim_order,
>> -                    highest_zoneidx);
>> +                    migratetype, highest_zoneidx);
>>             /* Read the new order and highest_zoneidx */
>>           alloc_order = READ_ONCE(pgdat->kswapd_order);
>> +        migratetype = kswapd_migratetype(pgdat, migratetype);
>>           highest_zoneidx = kswapd_highest_zoneidx(pgdat,
>>                               highest_zoneidx);
>>           WRITE_ONCE(pgdat->kswapd_order, 0);
>> +        WRITE_ONCE(pgdat->kswapd_migratetype, MIGRATE_TYPES);
>>           WRITE_ONCE(pgdat->kswapd_highest_zoneidx, MAX_NR_ZONES);
>>             ret = try_to_freeze();
>> @@ -3934,8 +3970,8 @@ static int kswapd(void *p)
>>            * request (alloc_order).
>>            */
>>           trace_mm_vmscan_kswapd_wake(pgdat->node_id, highest_zoneidx,
>> -                        alloc_order);
>> -        reclaim_order = balance_pgdat(pgdat, alloc_order,
>> +                        alloc_order, migratetype);
>> +        reclaim_order = balance_pgdat(pgdat, alloc_order, migratetype,
>>                           highest_zoneidx);
>>           if (reclaim_order < alloc_order)
>>               goto kswapd_try_sleep;
>> @@ -3953,11 +3989,12 @@ static int kswapd(void *p)
>>    * has failed or is not needed, still wake up kcompactd if only 
>> compaction is
>>    * needed.
>>    */
>> -void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
>> +void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order, 
>> int migratetype
>>              enum zone_type highest_zoneidx)
>>   {
>>       pg_data_t *pgdat;
>>       enum zone_type curr_idx;
>> +    int curr_migratetype;
>>         if (!managed_zone(zone))
>>           return;
>> @@ -3967,6 +4004,7 @@ void wakeup_kswapd(struct zone *zone, gfp_t 
>> gfp_flags, int order,
>>         pgdat = zone->zone_pgdat;
>>       curr_idx = READ_ONCE(pgdat->kswapd_highest_zoneidx);
>> +    curr_migratetype = READ_ONCE(pgdat->kswapd_migratetype);
>>         if (curr_idx == MAX_NR_ZONES || curr_idx < highest_zoneidx)
>>           WRITE_ONCE(pgdat->kswapd_highest_zoneidx, highest_zoneidx);
>> @@ -3974,6 +4012,9 @@ void wakeup_kswapd(struct zone *zone, gfp_t 
>> gfp_flags, int order,
>>       if (READ_ONCE(pgdat->kswapd_order) < order)
>>           WRITE_ONCE(pgdat->kswapd_order, order);
>>   +    if (curr_migratetype == MIGRATE_TYPES || 
>> is_migrate_movable(migratetype))
>> +        WRITE_ONCE(pgdat->kswapd_migratetype, migratetype);
>> +
>>       if (!waitqueue_active(&pgdat->kswapd_wait))
>>           return;
>>   @@ -3994,7 +4035,7 @@ void wakeup_kswapd(struct zone *zone, gfp_t 
>> gfp_flags, int order,
>>       }
>>         trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, 
>> highest_zoneidx, order,
>> -                      gfp_flags);
>> +                      migratetype, gfp_flags);
>>       wake_up_interruptible(&pgdat->kswapd_wait);
>>   }
>>   @@ -4017,6 +4058,7 @@ unsigned long shrink_all_memory(unsigned long 
>> nr_to_reclaim)
>>           .may_writepage = 1,
>>           .may_unmap = 1,
>>           .may_swap = 1,
>> +        .may_cma = 1,
>>           .hibernation_mode = 1,
>>       };
>>       struct zonelist *zonelist = node_zonelist(numa_node_id(), 
>> sc.gfp_mask);
>> @@ -4176,6 +4218,7 @@ static int __node_reclaim(struct pglist_data 
>> *pgdat, gfp_t gfp_mask, unsigned in
>>           .may_writepage = !!(node_reclaim_mode & RECLAIM_WRITE),
>>           .may_unmap = !!(node_reclaim_mode & RECLAIM_UNMAP),
>>           .may_swap = 1,
>> +        .may_cma = movable_reclaim(gfp_mask),
>>           .reclaim_idx = gfp_zone(gfp_mask),
>>       };
>>
>
>