MVME5500 status, and on to PMC interrupts

[Peter Dufault]

I've made some progress with the MVME5500 and I'm starting to see 
what's up with the PCI stuff.

By using the "BSP_PCIxFindDevice" instead of BSP_PciFindDevice in the 
mvme5500 tree and keepintg track of which "pci_num" the PMC modules 
were found on, I could then use "PCIx_read_config_dword" and friends to 
get both of the two PMC boards mounted on the MVME5500 working, or at 
least decoded into I/O space and with assigned interrupts.

The PMC card on the other side of the DEC bridge on the PMC_SPAN 
daughter card seems to return data, but any attempts to access the 
returned address space crashes.  I assume the cleanup that needs to be 
done to the PCI code is to fold everything into the shared PCI 
directory and properly hand in that pci_num.  I still don't understand 
this completely, though, since the PCIx functions seem to have four 
values: PCI number, bus, slot, and function and not just the three I'd 
heard mentioned before.

I next tried to install interrupt handlers.  There are disconcerting 
Intel PIC comments in the interrupt handler code that should be cleaned 
up.  I used BSP_install_rtems_irq_handler to install what I thought 
would work, and caused the board to generate an interrupt, and I could 
see from its registers that it thought it was generating an interrupt, 
but nothing happened.  That's when I found out that the "Marvell 
GT64260 system controller" on the MVME5500 is undocumented, and that 
you need an NDA to get the documentation or even to talk to technical 
support at Marvell.

So if anyone knows how to enable the PMC interrupts on the MVME5500, 
I'd love to know.

Petr

Peter Dufault
HD Associates, Inc.