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:   Sat, 2 Mar 2019 17:32:11 -0500
From:   Dennis Zhou <dennis@kernel.org>
To:     Peng Fan <peng.fan@nxp.com>
Cc:     Tejun Heo <tj@kernel.org>, Christoph Lameter <cl@linux.com>,
        Vlad Buslov <vladbu@mellanox.com>,
        "kernel-team@fb.com" <kernel-team@fb.com>,
        "linux-mm@kvack.org" <linux-mm@kvack.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 04/12] percpu: manage chunks based on contig_bits instead
 of free_bytes
Message-ID: <20190302223211.GC1196@dennisz-mbp.home>
References: <20190228021839.55779-1-dennis@kernel.org>
 <20190228021839.55779-5-dennis@kernel.org>
 <AM0PR04MB4481BE90E46F3635B7131CED88770@AM0PR04MB4481.eurprd04.prod.outlook.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <AM0PR04MB4481BE90E46F3635B7131CED88770@AM0PR04MB4481.eurprd04.prod.outlook.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Mar 02, 2019 at 01:48:20PM +0000, Peng Fan wrote:
> 
> 
> > -----Original Message-----
> > From: owner-linux-mm@kvack.org [mailto:owner-linux-mm@kvack.org] On
> > Behalf Of Dennis Zhou
> > Sent: 2019年2月28日 10:19
> > To: Dennis Zhou <dennis@kernel.org>; Tejun Heo <tj@kernel.org>; Christoph
> > Lameter <cl@linux.com>
> > Cc: Vlad Buslov <vladbu@mellanox.com>; kernel-team@fb.com;
> > linux-mm@kvack.org; linux-kernel@vger.kernel.org
> > Subject: [PATCH 04/12] percpu: manage chunks based on contig_bits instead
> > of free_bytes
> > 
> > When a chunk becomes fragmented, it can end up having a large number of
> > small allocation areas free. The free_bytes sorting of chunks leads to
> > unnecessary checking of chunks that cannot satisfy the allocation.
> > Switch to contig_bits sorting to prevent scanning chunks that may not be able
> > to service the allocation request.
> > 
> > Signed-off-by: Dennis Zhou <dennis@kernel.org>
> > ---
> >  mm/percpu.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/mm/percpu.c b/mm/percpu.c
> > index b40112b2fc59..c996bcffbb2a 100644
> > --- a/mm/percpu.c
> > +++ b/mm/percpu.c
> > @@ -234,7 +234,7 @@ static int pcpu_chunk_slot(const struct pcpu_chunk
> > *chunk)
> >  	if (chunk->free_bytes < PCPU_MIN_ALLOC_SIZE || chunk->contig_bits
> > == 0)
> >  		return 0;
> > 
> > -	return pcpu_size_to_slot(chunk->free_bytes);
> > +	return pcpu_size_to_slot(chunk->contig_bits * PCPU_MIN_ALLOC_SIZE);
> >  }
> > 
> >  /* set the pointer to a chunk in a page struct */
> 
> Reviewed-by: Peng Fan <peng.fan@nxp.com>
> 
> Not relevant to this patch, another optimization to percpu might be good
> to use per chunk spin_lock, not gobal pcpu_lock.
> 

Percpu memory itself is expensive and for the most part shouldn't be
part of the critical path. Ideally, we don't have multiple chunks being
allocated simultaneously because once an allocation is given out, the
chunk is pinned until all allocations are freed.

Thanks,
Dennis