On Thu, Oct 19, 2017 at 5:16 AM, Christoph Hellwig <hch@xxxxxx> wrote:
> On Wed, Oct 18, 2017 at 10:41:17AM -0400, Jim Quinlan wrote:
>> That's what brcm_to_{pci,cpu} are for -- they keep a list of the
>> dma-ranges given in the PCIe DT node, and translate from system memory
>> addresses to pci-space addresses, and vice versa. As long as people
>> are using the DMA API it should work. It works for all of the ARM,
>> ARM64, and MIPS Broadcom systems I've tested, using eight different EP
>> devices. Note that I am not thrilled to be advocating this mechanism
>> but it seemed the best alternative.
>
> Say we are using your original example ranges:
>
> memc0-a@[ 0....3fffffff] <=> pci@[ 0....3fffffff]
> memc0-b@[100000000...13fffffff] <=> pci@[ 40000000....7fffffff]
> memc1-a@[ 40000000....7fffffff] <=> pci@[ 80000000....bfffffff]
> memc1-b@[300000000...33fffffff] <=> pci@[ c0000000....ffffffff]
> memc2-a@[ 80000000....bfffffff] <=> pci@[100000000...13fffffff]
> memc2-b@[c00000000...c3fffffff] <=> pci@[140000000...17fffffff]
>
> and now you get a dma mapping request for physical addresses
> 3fffff00 to 4000000f, which would span two of your ranges. How
> is this going to work?
The only way to prevent this is to reserve a single page at the end of
the first memory region of any pair that are adjacent in physical
memory. A hack, yes, but I don't see an easier way out of this. Many
if not most of our boards do not have adjacent regions and would not
need this.
Overriding phys_to_dma/dma_to_phys comes with the same overlap problem
(MIPS solution and possible ARM/ARM64 solution).
>
>> I would prefer that the same code work for all three architectures.
>> What I would like from ARM/ARM64 is the ability to override
>> phys_to_dma() and dma_to_phys(); I thought the chances of that being
>> accepted would be slim. But you are right, I should ask the
>> maintainers.
>
> It is still better than trying to stack dma ops, which is a receipe
> for problems down the road.
Let me send out V2 of my patchset and also send it to the ARM/ARM64
maintainers as you suggested; perhaps there is an alternative.
