Re: [PATCH][1/2] usb: dma bounce buffer support V3

"Magnus Damm" <magnus.damm@xxxxxxxxx> · Tue, 22 Jan 2008 15:50:58 +0900

On Jan 22, 2008 3:25 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> On Tue, 22 Jan 2008 15:12:53 +0900 "Magnus Damm" <magnus.damm@xxxxxxxxx> wrote:
>
> > On Jan 22, 2008 2:01 PM, Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> > > On Tue, 22 Jan 2008 12:20:10 +0900 Magnus Damm <magnus.damm@xxxxxxxxx> wrote:
> > >
> > > > usb: dma bounce buffer support V3
> > > >
> > > > This patch adds dma bounce buffer support to the usb core. These buffers
> > > > can be enabled with the HCD_LOCAL_MEM flag, and they make sure that all data
> > > > passed to the host controller is allocated using dma_alloc_coherent().
> > >
> > > It is unobvious (to me) why this code uses put_unaligned() and
> > > get_unaligned() in the places where it does.  So I'll assume that it will
> > > be unobvious to other reviewers and readers.
> >
> > The HCD_LOCAL_MEM dma bounce buffer code is using unaligned functions
> > to temporarily save a pointer at the end of the dma data buffer.
> > hcd_alloc_coherent() saves the pointer which is needed by
> > hcd_free_coherent(). The unaligned functions are used because the size
> > of the buffer may be uneven.
> >
> > It's of course possible to round up the size instead and use aligned
> > access if you prefer that. The memory area used for dma may be limited
> > though so I felt it was natural to optimize for small memory footprint
> > over access speed.
>
> that wasn't a patch ;)

Gotcha. Sorry about the attachment - I have no idea how to reply
inline and not mangle patches using gmail...

/ magnus