Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:2 as permitted sender) client-ip=2620:137:e000::3:2;
Date:   Thu, 14 Dec 2023 13:22:29 -0500
From:   Dan Schatzberg <schatzberg.dan@gmail.com>
To:     Michal Hocko <mhocko@suse.com>
Cc:     Johannes Weiner <hannes@cmpxchg.org>,
        Roman Gushchin <roman.gushchin@linux.dev>,
        Yosry Ahmed <yosryahmed@google.com>, Huan Yang <link@vivo.com>,
        linux-kernel@vger.kernel.org, cgroups@vger.kernel.org,
        linux-mm@kvack.org, Tejun Heo <tj@kernel.org>,
        Zefan Li <lizefan.x@bytedance.com>,
        Jonathan Corbet <corbet@lwn.net>,
        Shakeel Butt <shakeelb@google.com>,
        Muchun Song <muchun.song@linux.dev>,
        Andrew Morton <akpm@linux-foundation.org>,
        David Hildenbrand <david@redhat.com>,
        Matthew Wilcox <willy@infradead.org>,
        Kefeng Wang <wangkefeng.wang@huawei.com>,
        Yue Zhao <findns94@gmail.com>, Hugh Dickins <hughd@google.com>
Subject: Re: [PATCH V4 2/2] mm: add swapiness= arg to memory.reclaim
Message-ID: <ZXtH5T+/qs+dUqrz@dschatzberg-fedora-PF3DHTBV>
References: <20231213013807.897742-1-schatzberg.dan@gmail.com>
 <20231213013807.897742-3-schatzberg.dan@gmail.com>
 <ZXq_H4St_NSEFkcD@tiehlicka>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZXq_H4St_NSEFkcD@tiehlicka>
Precedence: bulk

On Thu, Dec 14, 2023 at 09:38:55AM +0100, Michal Hocko wrote:
> On Tue 12-12-23 17:38:03, Dan Schatzberg wrote:
> > Allow proactive reclaimers to submit an additional swappiness=<val>
> > argument to memory.reclaim. This overrides the global or per-memcg
> > swappiness setting for that reclaim attempt.
> 
> You are providing the usecase in the cover letter and Andrew usually
> appends that to the first patch in the series. I think it would be
> better to have the usecase described here.
> 
> [...]
> > @@ -1304,6 +1297,18 @@ PAGE_SIZE multiple when read back.
> >  	This means that the networking layer will not adapt based on
> >  	reclaim induced by memory.reclaim.
> >  
> > +The following nested keys are defined.
> > +
> > +	  ==========		================================
> > +	  swappiness		Swappiness value to reclaim with
> > +	  ==========		================================
> > +
> > +	Specifying a swappiness value instructs the kernel to perform
> > +	the reclaim with that swappiness value. Note that this has the
> > +	same semantics as the vm.swappiness sysctl - it sets the
> 
> same semantics as vm.swappiness applied to memcg reclaim with all the
> existing limitations and potential future extensions.

Thanks, will make this change.

> 
> > +	relative IO cost of reclaiming anon vs file memory but does
> > +	not allow for reclaiming specific amounts of anon or file memory.
> > +
> >    memory.peak
> >  	A read-only single value file which exists on non-root
> >  	cgroups.
> 
> The biggest problem with the implementation I can see, and others have
> pointed out the same, is how fragile this is. You really have to check
> the code and _every_ place which defines scan_control to learn that
> mem_cgroup_shrink_node, reclaim_clean_pages_from_list,
> reclaim_folio_list, lru_gen_seq_write, try_to_free_pages, balance_pgdat,
> shrink_all_memory and __node_reclaim. I have only checked couple of
> them, like direct reclaim and kswapd and none of them really sets up
> swappiness to the default memcg nor global value. So you effectively end
> up with swappiness == 0.
> 
> While the review can point those out it is quite easy to break and you
> will only learn about that very indirectly. I think it would be easier
> to review and maintain if you go with a pointer that would fallback to
> mem_cgroup_swappiness() if NULL which will be the case for every
> existing reclaimer except memory.reclaim with swappiness value.

I agree. My initial implementation used a pointer for this
reason. I'll switch this back. Just to be clear - I still need to
initialize scan_control.swappiness in all these other places right? It
appears to mostly be stack-initialized which means it has
indeterminate value, not necessarily zero.