Re: [PATCH] memcg, kmem: do not fail __GFP_NOFAIL charges

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 11 Sep 2019 14:00:02 +0200 Michal Hocko <mhocko@kernel.org> wrote:

> On Mon 09-09-19 13:22:45, Michal Hocko wrote:
> > On Fri 06-09-19 11:24:55, Shakeel Butt wrote:
> [...]
> > > I wonder what has changed since
> > > <http://lkml.kernel.org/r/20180525185501.82098-1-shakeelb@google.com/>.
> > 
> > I have completely forgot about that one. It seems that we have just
> > repeated the same discussion again. This time we have a poor user who
> > actually enabled the kmem limit.
> > 
> > I guess there was no real objection to the change back then. The primary
> > discussion revolved around the fact that the accounting will stay broken
> > even when this particular part was fixed. Considering this leads to easy
> > to trigger crash (with the limit enabled) then I guess we should just
> > make it less broken and backport to stable trees and have a serious
> > discussion about discontinuing of the limit. Start by simply failing to
> > set any limit in the current upstream kernels.
> 
> Any more concerns/objections to the patch? I can add a reference to your
> earlier post Shakeel if you want or to credit you the way you prefer.
> 
> Also are there any objections to start deprecating process of kmem
> limit? I would see it in two stages
> - 1st warn in the kernel log
> 	pr_warn("kmem.limit_in_bytes is deprecated and will be removed.
> 	        "Please report your usecase to linux-mm@kvack.org if you "
> 		"depend on this functionality."

pr_warn_once() :)

> - 2nd fail any write to kmem.limit_in_bytes
> - 3rd remove the control file completely

Sounds good to me.