Messages in this thread |  | | | Date | Tue, 23 Sep 2008 13:02:04 +0900 | | Subject | Re: [PATCH 0/3] fix GART to respect device's dma_mask about virtual mappings | | From | FUJITA Tomonori <> | |
On Mon, 22 Sep 2008 22:35:18 +0200
Andi Kleen <andi@firstfloor.org> wrote:
> On Tue, Sep 23, 2008 at 04:12:33AM +0900, FUJITA Tomonori wrote:
> > Sure, pci-nommu's map_sg/map_single doesn't handle it. But we handle
> > this issue somewhere else (like b44 keeps own DMA buffer)?
>
> Yes, b44 handles it on its own. It has to for 32bit which always
> has a nop map_sg/single. Also some other subsystems like the block layer
> do it implicitely.
Yeah, 'somewhere else' includes the block layer bounce.
> > 1. not all map_sg/map_single (e.g. pci-nommu) can't handle it.
>
> and especially i386
>
> > 2. we already have workarounds for it somewhere else so
> > map_sg/map_single don't need to handle it.
> > 3. I changed GART map_sg/map_signle to handle it. I thought if it
> > can handle it, for example, b44 doesn't go the workaround path. It
> > would be a good thing.
> > 4. But GART cannot remap to addresses < 4GB reliably, so my above
> > argument doesn't always work.
> > 5. Then my patch doesn't break anything but it's almost meaningless.
>
> Correct.
Thanks for the clarification.
|  |