video-buf warning

video-buf warning

[Meelis Roos]

This is 2.6.6-rc3 on a sparc64. The warning has been there for quite
some time, finally got around to report it.

  CC [M]  drivers/media/video/video-buf.o
drivers/media/video/video-buf.c: In function `videobuf_iolock':
drivers/media/video/video-buf.c:327: warning: cast from pointer to integer of different size

The specific code is
/* FIXME: need sanity checks for vb->boff */
bus   = (dma_addr_t)fbuf->base + vb->boff;

bus is dma_addr_t (==u32 on sparc64), base is void*

So if buf->base is really an arbitrary pointer, it might not fit into
u32. What is it actually?

-- 
Meelis Roos (mroos@linux.ee)

Re: video-buf warning

[David S. Miller]

On Wed, 28 Apr 2004 11:57:38 +0300 (EEST)
Meelis Roos <mroos@linux.ee> wrote:

>   CC [M]  drivers/media/video/video-buf.o
> drivers/media/video/video-buf.c: In function `videobuf_iolock':
> drivers/media/video/video-buf.c:327: warning: cast from pointer to integer of different size
 ...
> The specific code is
> /* FIXME: need sanity checks for vb->boff */
> bus   = (dma_addr_t)fbuf->base + vb->boff;
> 
> bus is dma_addr_t (==u32 on sparc64), base is void*
> 
> So if buf->base is really an arbitrary pointer, it might not fit into
> u32. What is it actually?

It is the physical address of a frame buffer, this code is allowing
user programs to point the video capturing to go directly onto a
frame buffer at a specific location.

Using dma_addr_t here is a poor choice, since that data type is to
be used to PCI DMA API interfaces for doing transfers to/from real
memory, not frame buffers and the like :-)

Unfortunately, there is no portable interface available for what
this code wants to do, which is device<-->device DMA transfers.
We would need to create an interface that took two device structures,
and some base+offset values, in order to provide a portable way to
do this, then a reworked version of this user interface would be needed
as well.

In short, it's long term to fix this up, don't worry about it for
now.