RE: Pegasos interrupt mapping and dmamem_alloc

[Tim Rightnour <root%garbled.net@localhost>]

On 03-Nov-2007 Frank Wille wrote:
> At least the last method for "super-annoying OFW" is affected, which reads
> the interrupt lines directly from the PCI bus. Reading the "interrupt"
> property returns a pin number from 1 to 4, but this is erroneously added to
> 'A' (generating pins from 'B' to 'E'!). The following diff fixes that:

Oops!

> panic: call to null-ptr from 0x2e03e8
> ---8<---
> 
> But then a new crash! The _dma_phys_to_bus_mem() function in the
> pci_bus_dma_tag structure was not initialized! I don't know why you are
> making a difference between pci_bus_dma_tag for macppc and for the rest?
> Usng the macppc-version instead, which initializes the two missing function
> pointers, fixes the problem:

Because the macppc one isn't correct.  They are using thier own internal
versions of the routines, as opposed to the generic powerpc ones.  I keep
forgetting to ask Macallan if they still need this hack or not. I suspect they
do not.

Initializing those is something that needs to be done on a per-port basis in
include/bus.h.  If you take a look at powerpc/include/prep_bus.h you can see
that we define macros for PHYS_TO_BUS_MEM and BUS_MEM_TO_PHYS.

It looks like macppc uses the generic ones, so we can probably just set the 
PHYS_TO_BUS_MEM and BUS_MEM_TO_PHYS macros in ofppc/include/bus.h to those
two functions, and things will be good to go.

> But also a new problem: now attaching ofb0 freezes the system... *sigh*

It's probably not worth fiddling with ofb.  I was talking to Macallan, and he
said macppc will be dumping that driver in favor of genfb.

---
Tim Rightnour <root%garbled.net@localhost>
NetBSD: Free multi-architecture OS http://www.netbsd.org/
Genecys: Open Source 3D MMORPG: http://www.genecys.org/