Re: [PATCH 7/9] mm/compaction: factor out code to test if we should run compaction for target order

[Kemeng Shi]

on 8/19/2023 8:27 PM, Baolin Wang wrote:
> 
> 
> On 8/15/2023 8:10 PM, Kemeng Shi wrote:
>>
>>
>> on 8/15/2023 4:53 PM, Baolin Wang wrote:
>>>
>>>
>>> On 8/5/2023 7:07 PM, Kemeng Shi wrote:
>>>> We always do zone_watermark_ok check and compaction_suitable check
>>>> together to test if compaction for target order should be runned.
>>>> Factor these code out for preparation to remove repeat code.
>>>>
>>>> Signed-off-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
>>>> ---
>>>>    mm/compaction.c | 42 +++++++++++++++++++++++++++++-------------
>>>>    1 file changed, 29 insertions(+), 13 deletions(-)
>>>>
>>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>>> index b5a699ed526b..26787ebb0297 100644
>>>> --- a/mm/compaction.c
>>>> +++ b/mm/compaction.c
>>>> @@ -2365,6 +2365,30 @@ bool compaction_zonelist_suitable(struct alloc_context *ac, int order,
>>>>        return false;
>>>>    }
>>>>    +/*
>>>> + * Should we do compaction for target allocation order.
>>>> + * Return COMPACT_SUCCESS if allocation for target order can be already
>>>> + * satisfied
>>>> + * Return COMPACT_SKIPPED if compaction for target order is likely to fail
>>>> + * Return COMPACT_CONTINUE if compaction for target order should be runned
>>>> + */
>>>> +static inline enum compact_result
>>>> +compaction_suit_allocation_order(struct zone *zone, unsigned int order,
>>>> +                 int highest_zoneidx, unsigned int alloc_flags)
>>>> +{
>>>> +    unsigned long watermark;
>>>> +
>>>> +    watermark = wmark_pages(zone, alloc_flags & ALLOC_WMARK_MASK);
>>>
>>> IIUC, the watermark used in patch 8 and patch 9 is different, right? Have you measured the impact of modifying this watermark?
>>>
>> Actually, there is no functional change intended. Consider wmark_pages with
>> alloc_flags = 0 is equivalent to min_wmark_pages, patch 8 and patch 9 still
>> use original watermark.
> 
> Can you use ALLOC_WMARK_MIN macro to make it more clear?
Sorry, I can't quite follow this. The watermark should differ with different
alloc_flags instead of WMARK_MIN hard-coded.
Patch 8 and patch 9 use watermark with WMARK_MIN as they get alloc_flags = 0.
> 
> And I think patch 8 and patch 9 should be squashed into patch 7 to convert all at once.
Sure, i could do this in next version.
> 
>>>> +    if (zone_watermark_ok(zone, order, watermark, highest_zoneidx,
>>>> +                  alloc_flags))
>>>> +        return COMPACT_SUCCESS;
>>>> +
>>>> +    if (!compaction_suitable(zone, order, highest_zoneidx))
>>>> +        return COMPACT_SKIPPED;
>>>> +
>>>> +    return COMPACT_CONTINUE;
>>>> +}
>>>> +
>>>>    static enum compact_result
>>>>    compact_zone(struct compact_control *cc, struct capture_control *capc)
>>>>    {
>>>> @@ -2390,19 +2414,11 @@ compact_zone(struct compact_control *cc, struct capture_control *capc)
>>>>        cc->migratetype = gfp_migratetype(cc->gfp_mask);
>>>>          if (compaction_with_allocation_order(cc->order)) {
>>>> -        unsigned long watermark;
>>>> -
>>>> -        /* Allocation can already succeed, nothing to do */
>>>> -        watermark = wmark_pages(cc->zone,
>>>> -                    cc->alloc_flags & ALLOC_WMARK_MASK);
>>>> -        if (zone_watermark_ok(cc->zone, cc->order, watermark,
>>>> -                      cc->highest_zoneidx, cc->alloc_flags))
>>>> -            return COMPACT_SUCCESS;
>>>> -
>>>> -        /* Compaction is likely to fail */
>>>> -        if (!compaction_suitable(cc->zone, cc->order,
>>>> -                     cc->highest_zoneidx))
>>>> -            return COMPACT_SKIPPED;
>>>> +        ret = compaction_suit_allocation_order(cc->zone, cc->order,
>>>> +                               cc->highest_zoneidx,
>>>> +                               cc->alloc_flags);
>>>> +        if (ret != COMPACT_CONTINUE)
>>>> +            return ret;
>>>>        }
>>>>          /*
>>>
>>>
> 
>