Re: [PATCH v4 1/4] mm/pageblock: mitigation cmpxchg false sharing in pageblock flags

[David Hildenbrand]

On 03.09.20 09:01, Alex Shi wrote:
> pageblock_flags is used as long, since every pageblock_flags is just 4
> bits, 'long' size will include 8(32bit machine) or 16 pageblocks' flags,
> that flag setting has to sync in cmpxchg with 7 or 15 other pageblock
> flags. It would cause long waiting for sync.
> 
> If we could change the pageblock_flags variable as char, we could use
> char size cmpxchg, which just sync up with 2 pageblock flags. it could
> relief the false sharing in cmpxchg.
> 
> Signed-off-by: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>
> Cc: linux-mm@xxxxxxxxx
> Cc: linux-kernel@xxxxxxxxxxxxxxx

Could you please

1. Send a cover letter and indicate the changees between versions. I
cannot find any in my mailbox or on -mm - is there any? (also, is there
a patch 4 ?)

2. Report proper performance numbers as requested, especially, over
multiple runs. This should go into patch 1/2. Are they buried somewhere?

3. Clarify what patch 3 does: Do we actually waste 8*sizeof(long) where
we only need 4 bits?

Also, breaking stuff in patch 1 and fixing it in patch 3 is not
acceptable. This breaks git bisect. Skimming over the patches I think
this is the case.

I am not convinced yet that we need and want this. As Alex mentioned, we
touch pageblock flags while holding the zone lock already in most cases
-  and as Mel mentiones, updates should be rare.

-- 
Thanks,

David / dhildenb