Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 26 Feb 2020 17:16:54 -0800
From:   Matthew Wilcox <willy@infradead.org>
To:     Yang Shi <yang.shi@linux.alibaba.com>
Cc:     Hugh Dickins <hughd@google.com>, kirill.shutemov@linux.intel.com,
        aarcange@redhat.com, akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [v2 PATCH] mm: shmem: allow split THP when truncating THP
 partially
Message-ID: <20200227011654.GF24185@bombadil.infradead.org>
References: <1575420174-19171-1-git-send-email-yang.shi@linux.alibaba.com>
 <alpine.LSU.2.11.1912041601270.12930@eggly.anvils>
 <00f0bb7d-3c25-a65f-ea94-3e2de8e9bcdd@linux.alibaba.com>
 <alpine.LSU.2.11.2002241831060.3084@eggly.anvils>
 <cba16817-8555-f84f-134a-1ff9f168247b@linux.alibaba.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <cba16817-8555-f84f-134a-1ff9f168247b@linux.alibaba.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Feb 26, 2020 at 09:43:53AM -0800, Yang Shi wrote:
> > No.  The pagevec_lookup_entries() calls from mm/truncate.c prefer the
> > new behavior - evicting the head from page cache removes all the tails
> > along with it, so getting the tails a waste of time there too, just as
> > it was in shmem_undo_range().
> 
> TBH I'm not a fun of this hack. This would bring in other confusion or
> complexity. Pagevec is supposed to count in the number of base page, now it
> would treat THP as one page, and there might be mixed base page and THP in
> one pagevec. But, I tend to agree avoiding getting those 14 extra pins at
> the first place might be a better approach. All the complexity are used to
> release those extra pins.

My long-term goal is to eradicate tail pages entirely, so a pagevec will
end up containing pages of different sizes.  If you want to help move
in this direction, I'd be awfully grateful.  But I wouldn't say that's
in any way a prerequisite for fixing this current problem.