Re: [PATCH 7/7] mm: compaction: Introduce sync-light migration foruse by compaction

[Mel Gorman]

On Wed, Nov 23, 2011 at 10:35:37PM +0800, Nai Xia wrote:
> On Wed, Nov 23, 2011 at 9:45 PM, Mel Gorman <mgorman@xxxxxxx> wrote:
> > On Wed, Nov 23, 2011 at 09:05:08PM +0800, Nai Xia wrote:
> >> > <SNIP>
> >> >
> >> > Where are you adding this check?
> >> >
> >> > If you mean in __unmap_and_move(), the check is unnecessary unless
> >> > another subsystem starts using sync-light compaction. With this series,
> >> > only direct compaction cares about MIGRATE_SYNC_LIGHT. If the page is
> >>
> >> But I am still a little bit confused that if MIGRATE_SYNC_LIGHT is only
> >> used by direct compaction and  another mode can be used by it:
> >> MIGRATE_ASYNC also does not write dirty pages, then why not also
> >> do an (current->flags & PF_MEMALLOC) test before writing out pages,
> >
> > Why would it be necessary?
> > Why would it be better than what is there now?
> 
> I mean, if
>    MIGRATE_SYNC_LIGHT --> (current->flags & PF_MEMALLOC) and
>    MIGRATE_SYNC_LIGHT --> no dirty writeback, and (current->flags & PF_MEMALLOC)
>                       --> (MIGRATE_SYNC_LIGHT || MIGRATE_ASYNC)
>    MIGRATE_ASYNC      --> no dirty writeback, then
> why not simply  (current->flags & PF_MEMALLOC) ---> no dirty writeback
> and keep the sync meaning as it was?
> 

Ok, I see what you mean. Instead of making MIGRATE_SYNC_LIGHT part of
the API, we could instead special case within migrate.c how to behave if
MIGRATE_SYNC && PF_MEMALLOC.

This would be functionally equivalent and satisfy THP users
but I do not see it as being easier to understand or easier
to maintain than updating the API. If someone in the future
wanted to use migration without significant stalls without
being PF_MEMALLOC, they would need to update the API like this.
There are no users like this today but automatic NUMA migration
might want to leverage something like MIGRATE_SYNC_LIGHT
(http://comments.gmane.org/gmane.linux.kernel.mm/70239)

-- 
Mel Gorman
SUSE Labs
--
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/