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;
From:   "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>
To:     Alexey Kardashevskiy <aik@ozlabs.ru>,
        Christoph Hellwig <hch@lst.de>
Cc:     iommu@lists.linux-foundation.org, linuxppc-dev@lists.ozlabs.org,
        Lu Baolu <baolu.lu@linux.intel.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Joerg Roedel <joro@8bytes.org>,
        Robin Murphy <robin.murphy@arm.com>,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/2] dma-mapping: add a dma_ops_bypass flag to struct
 device
In-Reply-To: <ffce1af6-a215-dee8-7b5c-2111f43accfd@ozlabs.ru>
References: <20200320141640.366360-1-hch@lst.de>
 <20200320141640.366360-2-hch@lst.de>
 <2f31d0dd-aa7e-8b76-c8a1-5759fda5afc9@ozlabs.ru>
 <20200323083705.GA31245@lst.de> <20200323085059.GA32528@lst.de>
 <87sghz2ibh.fsf@linux.ibm.com> <20200323172256.GB31269@lst.de>
 <ffce1af6-a215-dee8-7b5c-2111f43accfd@ozlabs.ru>
Date:   Tue, 24 Mar 2020 12:00:09 +0530
Message-ID: <87pnd22rke.fsf@linux.ibm.com>
MIME-Version: 1.0
Content-Type: text/plain
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Alexey Kardashevskiy <aik@ozlabs.ru> writes:

> On 24/03/2020 04:22, Christoph Hellwig wrote:
>> On Mon, Mar 23, 2020 at 09:07:38PM +0530, Aneesh Kumar K.V wrote:
>>>
>>> This is what I was trying, but considering I am new to DMA subsystem, I
>>> am not sure I got all the details correct. The idea is to look at the
>>> cpu addr and see if that can be used in direct map fashion(is
>>> bus_dma_limit the right restriction here?) if not fallback to dynamic
>>> IOMMU mapping.
>> 
>> I don't think we can throw all these complications into the dma
>> mapping code.  At some point I also wonder what the point is,
>> especially for scatterlist mappings, where the iommu can coalesce.
>
> This is for persistent memory which you can DMA to/from but yet it does
> not appear in the system as a normal memory and therefore requires
> special handling anyway (O_DIRECT or DAX, I do not know the exact
> mechanics). All other devices in the system should just run as usual,
> i.e. use 1:1 mapping if possible.

This is O_DIRECT with a user buffer that is actually mmap from a dax
mounted file system.

What we really need is something that will falback to iommu_map_page
based on dma_addr. ie. Something equivalent to current
dma_direct_map_page(), but instead of fallback to swiotlb_map page we
should fallback to iommu_map_page().

Something like?

dma_addr_t dma_direct_map_page(struct device *dev, struct page *page,
		unsigned long offset, size_t size, enum dma_data_direction dir,
		unsigned long attrs)
{
	phys_addr_t phys = page_to_phys(page) + offset;
	dma_addr_t dma_addr = phys_to_dma(dev, phys);

	if (unlikely(!dma_capable(dev, dma_addr, size, true))) {
			return iommu_map(dev, phys, size, dir, attrs);

		return DMA_MAPPING_ERROR;
	}

....
...


-aneesh