On Wed, Jun 05, 2002 at 03:47:47PM -0700, David S. Miller wrote:
> From: William Jhun <wjhun@ayrnetworks.com>
> Date: Wed, 5 Jun 2002 13:12:31 -0700
>
> In the current linux-mips implementation, this has some subtle problems:
> pci_unmap_{single,sg}() is essentially a no-op.
>
> Right, I see the problem. I'll think about this some more.
>
> As it stands now, I think the correct solution is to require
> pci_dma_prep_single() before giving the buffer back to the
> device after the read.
Right, that's what I was thinking. Is it asking a lot to demand that all
existing drivers that use this interface add pci_dma_prep_single()? How
will backward compatiblility with older drivers work? That's why I
suggested leaving pci_dma_sync_single() and adding
pci_dma_release_single() which can leave the cache flush to
pci_dma_prep_single(). It seems like elsewhere, like the D-cache
flushing interface (for virtual aliasing), both old and new interfaces
co-exist. Does this seem to work out O.K. in your experience?
Will
