Re: [PATCH v2 25/26] omap3isp: Move to videobuf2

[Tim Nordell <tim.nordell@logicpd.com>]

Laurent -

On 03/18/15 10:21, Laurent Pinchart wrote:
> Correct. sg_dma_address() should contain the DMA virtual address when 
> an IOMMU is used.
>> I was assuming it was checking the physical memory layout to it without
>> looking too closely to the code.  Armed with that knowledge, I'll dig a
>> little deeper to see if I can figure out what happened here.
I added a bit of code to where it was printing out the error code to 
print out the mapping as vb2_dc_get_contiguous_size(...) sees it.  I get 
back the following:

[  200.031249] sgt[0].addr = 0x400a0000 - 0x400affff
[  200.031280] sgt[1].addr = 0x400b0000 - 0x400bffff
[  200.031311] sgt[2].addr = 0x400c0000 - 0x400cffff
[  200.031341] sgt[3].addr = 0x400d0000 - 0x400dffff
[  200.031372] sgt[4].addr = 0x400e0000 - 0x400effff
[  200.031402] sgt[5].addr = 0x400f0000 - 0x400fffff
[  200.031433] sgt[6].addr = 0x40100000 - 0x4010ffff
[  200.031463] sgt[7].addr = 0x40110000 - 0x4011ffff
[  200.031494] sgt[8].addr = 0x40120000 - 0x4012ffff
[  200.031524] sgt[9].addr = 0x40098000 - 0x4009dfff
[  200.031524] contiguous mapping is too small 589824/614400

Notice that the last section is completely off the wall compared to the 
rest?

Digging through to find who is responsible for assigning the virtual 
addresses, I find that it's buried inside 
arch/arm/mm/dma-mapping.c:__alloc_iova(...).  This call is called 
individually for each entry in the scatter-gather table via 
__map_sg_chunk from iommu_map_sg(...).  If this is supposed to allocate 
a contiguous virtual memory region, it seems that __iommu_map_sg(...) 
should be considering the full buffer range rather than parts of the 
buffer at a time for the virtual allocation, similar to how 
__iommu_create_mapping(...) works in the same file.

- Tim