RE: [PATCH 00/14] DMA-mapping framework redesign preparation

[Marek Szyprowski <m.szyprowski@samsung.com>]

Hello,

On Friday, December 23, 2011 5:35 PM Matthew Wilcox wrote:

> On Fri, Dec 23, 2011 at 01:27:19PM +0100, Marek Szyprowski wrote:
> > The first issue we identified is the fact that on some platform (again,
> > mainly ARM) there are several functions for allocating DMA buffers:
> > dma_alloc_coherent, dma_alloc_writecombine and dma_alloc_noncoherent
> 
> Is this write-combining from the point of view of the device (ie iommu),
> or from the point of view of the CPU, or both?

It is about write-combining from the CPU point of view. Right now there are
no devices with such advanced memory interface to do write combining on the
DMA side, but I believe that they might appear at some point in the future 
as well.

> > The next step in dma mapping framework update is the introduction of
> > dma_mmap/dma_mmap_attrs() function. There are a number of drivers
> > (mainly V4L2 and ALSA) that only exports the DMA buffers to user space.
> > Creating a userspace mapping with correct page attributes is not an easy
> > task for the driver. Also the DMA-mapping framework is the only place
> > where the complete information about the allocated pages is available,
> > especially if the implementation uses IOMMU controller to provide a
> > contiguous buffer in DMA address space which is scattered in physical
> > memory space.
> 
> Surely we only need a helper which drivrs can call from their mmap routine
> to solve this?

On ARM architecture it is already implemented this way and a bunch of drivers
use dma_mmap_coherent/dma_mmap_writecombine calls. We would like to standardize
these calls across all architectures.

> > Usually these drivers don't touch the buffer data at all, so the mapping
> > in kernel virtual address space is not needed. We can introduce
> > DMA_ATTRIB_NO_KERNEL_MAPPING attribute which lets kernel to skip/ignore
> > creation of kernel virtual mapping. This way we can save previous
> > vmalloc area and simply some mapping operation on a few architectures.
> 
> I really think this wants to be a separate function.  dma_alloc_coherent
> is for allocating memory to be shared between the kernel and a driver;
> we already have dma_map_sg for mapping userspace I/O as an alternative
> interface.  This feels like it's something different again rather than
> an option to dma_alloc_coherent.

That is just a starting point for the discussion. 

I thought about this API a bit and came to conclusion that there is no much
difference between a dma_alloc_coherent which creates a mapping in kernel
virtual space and the one that does not. It is just a hint from the driver
that it will not use that mapping at all. Of course this attribute makes sense
only together with adding a dma_mmap_attrs() call, because otherwise drivers
won't be able to get access to the buffer data.

On coherent architectures where dma_alloc_coherent is just a simple wrapper
around alloc_pages_exact() such attribute can be simply ignored without any
impact on the drivers (that's the main idea behind dma attributes!).
However such hint will help a lot on non-coherent architectures where 
additional work need to be done to provide a cohenent mapping in kernel 
address space. It also saves some precious kernel resources like vmalloc
address range.

Best regards
-- 
Marek Szyprowski
Samsung Poland R&D Center