Re: [PATCH 21/48] writeback: make backing_dev_info host cgroup-specific bdi_writebacks

[Tejun Heo]

Hello Vivek.

On Fri, Mar 27, 2015 at 05:06:13PM -0400, Vivek Goyal wrote:
> I was curious to know that why do we need this "struct page *page" when
> trying to attach a inode to a bdi_writeback. Is using current's cgroup
> always not sufficient?

So, memcg's page ownership is first-use based and it never gets
updated once set till the page is released which means that there can
be corner cases where an inode is mostly faulted in by one cgroup and
then constantly dirtied by another.  Because the ownership belongs to
the initial cgroup which instantiated those pages, cgroup writeback
ends up considering the pages as belonging to that initial cgroup and
the foreign detection will trigger if it's being written by a
different cgroup.  Hmmmm... this isn't a huge problem as once the
foreign detection triggers, the problem will be corrected but still
when the page is availalbe, I think it makes sense to attach to the
page as that's what actually defines the ownership.

Thanks.

-- 
tejun
--
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/