Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 30 Mar 2022 22:14:22 -0700 (PDT)
From:   Hugh Dickins <hughd@google.com>
To:     Matthew Wilcox <willy@infradead.org>
cc:     Hugh Dickins <hughd@google.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Muchun Song <songmuchun@bytedance.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: WARNING: mm/workingset.c:567 shadow_lru_isolate
In-Reply-To: <YkTPUmMitEDHfjeH@casper.infradead.org>
Message-ID: <20f5a12f-416b-458b-80f1-e4c13cd1ec64@google.com>
References: <55db706b-9bce-b820-7d88-6392374cd4fc@google.com> <YkTPUmMitEDHfjeH@casper.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Precedence: bulk

On Wed, 30 Mar 2022, Matthew Wilcox wrote:
> On Mon, Mar 28, 2022 at 09:24:00PM -0700, Hugh Dickins wrote:
> > I'm not familiar with workingset.c, I'll make no guesses, over to you!
> 
> I haven't been able to reproduce it, but I applied Feynman's Problem
> Solving Algorithm, and I think this will do the trick:

Excellent! Yes, this fixes it - thanks.

Hugh

> 
> diff --git a/lib/xarray.c b/lib/xarray.c
> index b95e92598b9c..d3b168f619b1 100644
> --- a/lib/xarray.c
> +++ b/lib/xarray.c
> @@ -1079,6 +1079,7 @@ void xas_split(struct xa_state *xas, void *entry, unsigned int order)
>  					xa_mk_node(child));
>  			if (xa_is_value(curr))
>  				values--;
> +			xas_update(xas, child);
>  		} else {
>  			unsigned int canon = offset - xas->xa_sibs;
>  
> @@ -1093,6 +1094,7 @@ void xas_split(struct xa_state *xas, void *entry, unsigned int order)
>  	} while (offset-- > xas->xa_offset);
>  
>  	node->nr_values += values;
> +	xas_update(xas, node);
>  }
>  EXPORT_SYMBOL_GPL(xas_split);
>  #endif
> 
> (the key insight is that the update function should have removed the
> node from the list; the WARN_ON really means "this node should not be on
> the list", so there's somewhere that we're forgetting to call
> xas_update(), or somewhere that we're forgetting to call
> mapping_set_update().  I didn't find any of the latter, but I did find
> one of the former)