Re: [PATCH 4/8] mm/compaction: remove stale fast_find_block flag in isolate_migratepages

[Kemeng Shi]

on 8/1/2023 11:34 AM, Baolin Wang wrote:
> 
> 
> On 8/1/2023 11:24 AM, Kemeng Shi wrote:
>>
>>
>> on 8/1/2023 10:42 AM, Baolin Wang wrote:
>>>
>>>
>>> On 7/29/2023 1:10 AM, Kemeng Shi wrote:
>>>> In old code, we set skip to found page block in fast_find_migrateblock. So
>>>> we use fast_find_block to avoid skip found page block from
>>>> fast_find_migrateblock.
>>>> In 90ed667c03fe5 ("Revert "Revert "mm/compaction: fix set skip in
>>>> fast_find_migrateblock"""), we remove skip set in fast_find_migrateblock,
>>>> then fast_find_block is useless.
>>>>
>>>> Signed-off-by: Kemeng Shi <shikemeng@xxxxxxxxxxxxxxx>
>>>> ---
>>>>    mm/compaction.c | 12 +-----------
>>>>    1 file changed, 1 insertion(+), 11 deletions(-)
>>>>
>>>> diff --git a/mm/compaction.c b/mm/compaction.c
>>>> index ad535f880c70..09c36251c613 100644
>>>> --- a/mm/compaction.c
>>>> +++ b/mm/compaction.c
>>>> @@ -1949,7 +1949,6 @@ static isolate_migrate_t isolate_migratepages(struct compact_control *cc)
>>>>        const isolate_mode_t isolate_mode =
>>>>            (sysctl_compact_unevictable_allowed ? ISOLATE_UNEVICTABLE : 0) |
>>>>            (cc->mode != MIGRATE_SYNC ? ISOLATE_ASYNC_MIGRATE : 0);
>>>> -    bool fast_find_block;
>>>>          /*
>>>>         * Start at where we last stopped, or beginning of the zone as
>>>> @@ -1961,13 +1960,6 @@ static isolate_migrate_t isolate_migratepages(struct compact_control *cc)
>>>>        if (block_start_pfn < cc->zone->zone_start_pfn)
>>>>            block_start_pfn = cc->zone->zone_start_pfn;
>>>>    -    /*
>>>> -     * fast_find_migrateblock marks a pageblock skipped so to avoid
>>>> -     * the isolation_suitable check below, check whether the fast
>>>> -     * search was successful.
>>>> -     */
>>>> -    fast_find_block = low_pfn != cc->migrate_pfn && !cc->fast_search_fail;
>>>> -
>>>>        /* Only scan within a pageblock boundary */
>>>>        block_end_pfn = pageblock_end_pfn(low_pfn);
>>>>    @@ -1976,7 +1968,6 @@ static isolate_migrate_t isolate_migratepages(struct compact_control *cc)
>>>>         * Do not cross the free scanner.
>>>>         */
>>>>        for (; block_end_pfn <= cc->free_pfn;
>>>> -            fast_find_block = false,
>>>>                cc->migrate_pfn = low_pfn = block_end_pfn,
>>>>                block_start_pfn = block_end_pfn,
>>>>                block_end_pfn += pageblock_nr_pages) {
>>>> @@ -2007,8 +1998,7 @@ static isolate_migrate_t isolate_migratepages(struct compact_control *cc)
>>>>             * before making it "skip" so other compaction instances do
>>>>             * not scan the same block.
>>>>             */
>>>> -        if (pageblock_aligned(low_pfn) &&
>>>> -            !fast_find_block && !isolation_suitable(cc, page))
>>>> +        if (pageblock_aligned(low_pfn) && !isolation_suitable(cc, page))
>>>
>>> I do not think so. If the pageblock is found by fast_find_migrateblock(), that means it definitely has not been set the skip flag, so there is not need to call isolation_suitable() if fast_find_block is true, right?
>>>
>>>
>> Actually, found pageblock could be set skip as:
>> 1. other compactor could mark this pageblock as skip after zone lock is realeased
>> in fast_find_migrateblock.
> 
> Yes, but your patch also can not close this race window, that means it can also be set skip flag after the isolation_suitable() validation by other compactors.
> 
Yes, I think it's still worth to remove a lot of fast_find_block relevant check and reduce
code complexity with one redundant isolation_suitable which may skip some block with luck.
>> 2. fast_find_migrateblock may uses pfn from reinit_migrate_pfn which is previously found
>> and sacnned. It could be fully sacnned and marked skip after it's first return from
> 
> Right, but now the 'fast_find_block' is false, and we will call isolation_suitable() to validate the skip flag.
> 
Right, sorry for missing that.

But it's ok to keep the fast_find_block if you insist and I will just correct the stale
comment that "fast_find_migrateblock marks a pageblock skipped ..." in next version.
Thanks!
>> fast_find_migrateblock and it should be skipped.
>> Thanks!
> 

-- 
Best wishes
Kemeng Shi