Re: [Patch] DMA mapping API for Alpha

[Marc Zyngier <mzyngier@freesurf.fr>]

>>>>> "Ivan" == Ivan Kokshaysky <ink@jurassic.park.msu.ru> writes:

Ivan> Agreed, but what if your EISA-PCI bridge has only 30 address
Ivan> lines wired to PCI? Yes, we can check this for EISA device
Ivan> because it has *real* PCI parent (thanks, Marc :-), but what
Ivan> about ISA/legacy/whatever drivers?  I doubt that all of them
Ivan> bother to set dma_mask pointer (so you can have an oops there).

If the driver is not ported to the device API, than we pass NULL as a
device pointer, and then we fallback to the old behaviour, aka
dma_mask=0x00ffffff. If the driver supplies a dev pointer, but fails
to set its dma_mask pointer, than it is a driver bug that should be
fixed.

And yes, the EISA subsystem should properly report the dma_mask to
attached devices (patches for that are in mm tree, and sent to Linus).

        M.
-- 
Places change, faces change. Life is so very strange.