svn commit: r192026 - head/share/man/man9

[John Baldwin]

On Tuesday 12 May 2009 5:28:13 pm Marius Strobl wrote:
> On Tue, May 12, 2009 at 04:13:06PM -0500, Robert Noland wrote:
> > On Tue, 2009-05-12 at 20:56 +0000, Marius Strobl wrote:
> > > Author: marius
> > > Date: Tue May 12 20:56:34 2009
> > > New Revision: 192026
> > > URL: http://svn.freebsd.org/changeset/base/192026
> > > 
> > > Log:
> > >   Correct r190283 (partially reverting it) as on sparc64 BUS_DMA_NOCACHE
> > >   actually is only valid for bus_dmamap_load().
> > 
> > Ok, this is getting very confusing...  This means that code has to set
> > this flag on both alloc and load to allow for somethine resembling
> > consistent behavior.
> > 
> 
> Personally I don't understand why amd64 and i386 where decided to
> implement BUS_DMA_NOCACHE for bus_dmamem_alloc(9) only as this
> is less flexible than using it with bus_dmamap_load(9) (which also
> is the older existing implementation). Anyway, <sys/bus_dma.h>
> documents BUS_DMA_NOCACHE and BUS_DMA_NOWRITE as "non-standard or
> specific to only certain architectures" so I think it's okay for
> the usage of these flags to differ across them.

Hmm, it might be possible to move it to bus_dmamap_load() instead.

-- 
John Baldwin