Re: rtVAX300 .. need help..

[Roger Ivie <rivie%ridgenet.net@localhost>]

On Tue, 8 Jan 2013, Holm Tiffe wrote:
> I know you are all here looking what happens but "without a little help
> from my friends.."

Well, speaking only for myself, I'm pretty much a NetBSD dilettante.
I've been using it for *ages*, but I've never really done any kernel
hacking on it.

I've done plenty of work with raw VAX hardware and with VMS, but all my
VAX related work was done a long, long time ago. The memory fades.

> Are there further hints what I should do with this VS_REGS mapping?
> Is it normal for an VAX CPU or for the CVAX to have memory mapped
> registers? Why are VAX49 and VAX53 are using an different address?

Addresses are typically whatever happened to be convenient for the
system designer. For busless systems, there weren't a lot of conventions
followed.

I stumbled onto an example of these differences between the VAXstation
4000/60 (I suppose that'd be a KA46) and the /90 when I was working up
some code to bit-bang a serial console intended for a little NVAX
processor module designed to fit into the CPU socket of an AlphaStation
250mumble. I prototyped the thing on the /60 using the modem control
lines of the communications port. When I tried it on the /90, I
discovered A) that the flash on the /90 occupied the space taken by the
console on the /60 and that B) fiddling with the modem control signals
at the /60 address erased the flash on a /90.

Of course, what really got me in trouble was fixing it. Since the flash
was really close to the power connector on the /90s, we installed
sockets. Turned out the machine was on a maintenance contract and Field
Circus was Not Amused.

> #if VAX49 || VAX53
>        case VAX_BTYP_53:
>        case VAX_BTYP_49:
>                sc->sc_vsregs = vax_map_physmem(0x25c00000, 1);
>                sc->sc_intreq = (char *)sc->sc_vsregs + 12;
>                sc->sc_intclr = (char *)sc->sc_vsregs + 12;
>                sc->sc_intmsk = (char *)sc->sc_vsregs + 8;
>                vsbus_dma_init(sc, 8192);
>                break;
> #endif

I don't have any documentation on the /90, but one thing I discovered
when I designed the TURBOchannel adapter for the /60 (which was also
used in the /90) was that the /90 uses edge-triggered interrupts instead
of level-triggered. I worked around that in my own VMS device drivers by
fiddling with an interrupt enable bit in the TURBOchannel adapter (after
servicing an interrupt, I would clear and set the interrupt enable so
that a new edge would be created on a /90 if the interrupt was still
asserted).

These registers are presumably used for that sort of interrupt
management on those platforms.

These registers would wind up in whatever FPGA they designed to support
the system. These would be system-specific, so they would very from
system to system.
--
roger ivie
rivie%ridgenet.net@localhost