Subject: Re: [PATCH 27/33] dma-direct: use node local allocations for coherent
 memory
To: Christoph Hellwig <hch@lst.de>
Cc: iommu@lists.linux-foundation.org, linux-mips@linux-mips.org,
        linux-ia64@vger.kernel.org, linux-sh@vger.kernel.org,
        sparclinux@vger.kernel.org, Guan Xuetao <gxt@mprc.pku.edu.cn>,
        linux-arch@vger.kernel.org, linux-s390@vger.kernel.org,
        linux-c6x-dev@linux-c6x.org, linux-hexagon@vger.kernel.org,
        x86@kernel.org, Konrad Rzeszutek Wilk <konrad@darnok.org>,
        linux-snps-arc@lists.infradead.org,
        linux-m68k@lists.linux-m68k.org, patches@groups.riscv.org,
        linux-metag@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        Michal Simek <monstr@monstr.eu>, linux-parisc@vger.kernel.org,
        linux-cris-kernel@axis.com, linux-kernel@vger.kernel.org,
        linux-alpha@vger.kernel.org, linuxppc-dev@lists.ozlabs.org
References: <20180110080027.13879-1-hch@lst.de>
 <20180110080027.13879-28-hch@lst.de>
 <3672aa56-b85c-5d2c-0c0e-709031b0c0a0@arm.com>
 <20180110153017.GD17790@lst.de>
From: Robin Murphy <robin.murphy@arm.com>
Message-ID: <3f37ba8d-66fa-80e2-bbe1-77c2a4323d6b@arm.com>
Date: Wed, 10 Jan 2018 16:49:58 +0000
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.5.0
MIME-Version: 1.0
In-Reply-To: <20180110153017.GD17790@lst.de>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-GB
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org

On 10/01/18 15:30, Christoph Hellwig wrote:
> On Wed, Jan 10, 2018 at 12:06:22PM +0000, Robin Murphy wrote:
>> On 10/01/18 08:00, Christoph Hellwig wrote:
>>> To preserve the x86 behavior.
>>
>> And combined with patch 10/22 of the SWIOTLB refactoring, this means
>> SWIOTLB allocations will also end up NUMA-aware, right? Great, that's what
>> we want on arm64 too :)
> 
> Well, only for swiotlb allocations that can be satisfied by
> dma_direct_alloc.  If we actually have to fall back to the swiotlb
> buffers there is not node affinity yet.

Yeah, when I looked into it I reached the conclusion that per-node 
bounce buffers probably weren't worth it - if you have to bounce you've 
already pretty much lost the performance game, and if the CPU doing the 
bouncing happens to be on a different node from the device you've 
certainly lost either way. Per-node CMA zones we definitely *would* 
like, but that's a future problem (it looks technically feasible without 
huge infrastructure changes, but fiddly).

Robin.