[parisc-linux] Re: CCIO dma io_command and related io_tlb format questions.

[Grant Grundler]

On Sat, Oct 14, 2006 at 11:35:47PM +0000, Joel Soete wrote:
...
> So here, I guess that software reflect this hw configuration and when a 
> lasi module driver (let say e.g. 53c710) need some io resources it would 
> forward the request to ccio module driver which would most probably have to 
> use its own sg list management functions (like ccio_map_single(), 
> ccio_unmap_single(), ccio_map_sg(), ccio_unmap_sg(), etc) which I called 
> "sg list management in ccio".

This is correct for C/K/D/R class machines which have U2/Uturn chip.

> Otc on b180 lasi module would better have to address there request to the 
> "Pseudo Bus Converter", i.e. quasi directly.

"Pseudo" means "fake". Ie the "Pseudo bus converter" just a place holder
in the HW path on B-class machines and not a real piece of HW.

> So why/how the kernel on b180 would use those same functions 
> (ccio_map_single(), ccio_unmap_single(), ccio_map_sg(), ccio_unmap_sg(), 
> etc) to manage sg list of a non-exitent device???

See arch/parisc/kernel/pci-dma.c for B180 DMA mapping functions.

> What did I miss?

You missed where/how dma_ops gets assigned.
Find that and I think you'll understand one more bit of the puzzle.

grant