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;
MIME-Version: 1.0
In-Reply-To: <alpine.DEB.2.10.1810132256240.4908@lmark-linux.qualcomm.com>
References: <1539214413-26173-1-git-send-email-john.stultz@linaro.org>
 <7534ca1d-f874-7809-6125-d9fc72f70e39@redhat.com> <alpine.DEB.2.10.1810132256240.4908@lmark-linux.qualcomm.com>
From:   John Stultz <john.stultz@linaro.org>
Date:   Mon, 15 Oct 2018 09:31:32 -0700
Message-ID: <CALAqxLWw0k7_cB2hY=2gPbYX7omKob7eV9HJizxYCVc=n=K9EA@mail.gmail.com>
Subject: Re: [PATCH] staging: ion: Rework ion_map_dma_buf() to minimize re-mapping
To:     Liam Mark <lmark@codeaurora.org>
Cc:     Laura Abbott <labbott@redhat.com>,
        lkml <linux-kernel@vger.kernel.org>,
        Beata Michalska <Beata.Michalska@arm.com>,
        Matt Szczesiak <matt.szczesiak@arm.com>,
        Anders Pedersen <Anders.Pedersen@arm.com>,
        John Reitan <John.Reitan@arm.com>,
        Sumit Semwal <sumit.semwal@linaro.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Todd Kjos <tkjos@android.com>,
        Martijn Coenen <maco@android.com>,
        dri-devel <dri-devel@lists.freedesktop.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Oct 13, 2018 at 11:01 PM, Liam Mark <lmark@codeaurora.org> wrote:
> On Fri, 12 Oct 2018, Laura Abbott wrote:
>> I thought there might have been some Qualcomm
>> stuff that did that (Liam? Todd?)
>
> Yes we have a form of "lazy mapping", which clients can opt into using,
> which results in iommu page table mapping not being unmaped on dma-unamp.
> Instead they are unmapped when the buffer is destroyed.
>
> It is important to note that in our "lazy mapping" implementation cache
> maintenance is still applied on dma-map and dma-unmap.
> If you need a full description of this feature I can provide it.
>
>>
>> I suspect most of the cost of the dma_map/dma_unmap is from the
>> cache flushing and not the actual mapping operations. If this
>> is the case, another option might be to figure out how to
>> incorporate dma_attrs so drivers can use DMA_ATTR_SKIP_CPU_SYNC
>> to decide when they actually want to sync.
>
> We have support for this locally on our 4.14 branch.
>
> We have added a dma_map_attrs field to the dma_buf_attachment struct,
> clients can then specify dma-attributes here such as
> DMA_ATTR_SKIP_CPU_SYNC before dma-mapping the buffer, then we ensure that
> these dma attributes are passed to dma_map_sg_attrs when ion_map_dma_buf
> is called (same for ion_unmap_dma_buf).

Do you happen to have a pointer to the tree if its public?

thanks
-john