NetBSD/sys/arch/sandpoint
jdolecek c82ab2eb79 now that mem_no is emitted by config(8), there is no reason to keep
copy of more or less identical iskmemdev() for every arch; move the function
to spec_vnop.c, and g/c machine-dependant copies
2002-10-26 13:50:17 +00:00
..
compile
conf now that mem_no is emitted by config(8), there is no reason to keep 2002-10-26 13:50:17 +00:00
include Merge the gehenna-devsw branch into the trunk. 2002-09-06 13:18:43 +00:00
isa Remove unreferenced file. 2002-05-04 18:23:35 +00:00
pci Add trailing ; to CFATTACH_DECL. 2002-10-02 15:52:22 +00:00
sandpoint now that mem_no is emitted by config(8), there is no reason to keep 2002-10-26 13:50:17 +00:00
Makefile
README Sync with reality. Add NetBSD tag. 2001-03-06 20:16:09 +00:00

/*	$NetBSD: README,v 1.2 2001/03/06 20:16:09 briggs Exp $	*/

Overview

This is a port to the Motorola "SandPoint" evaluation system.  The
SandPoint is the successor to the "Yellowknife" system.  The system
can be fitted with different PMCs (Processor Mezzanine Cards).  This
port is specifically for the rev X2 motherboard system with the PPC
8240 PMC rev X4 installed.  It also works with the Altimus X2 PMC
(MPC7400 with MPC107).

All references (cf) listed here are for the MPC8240 Integrated Processor
User's Manual.

Information on the Sandpoint can be found on Motorola's web site:
http://www.mot.com/SPS/PowerPC/teksupport/refdesigns/sandpoint.html


SandPoint Hardware Configuration

This port was developed on a Sandpoint X2 motherboard with a Unity X4 PMC.

This port assumes that the jumpers are set as follows:
	S3/S4	- Mode 1: PMC w/o IDE (switches opposite, one nearest PCI
		  slot toward near edge)
	S5	- Interrupt to PMC normal (switch toward near edge)
	S6	- Local I/O shared with slot 2 (switch toward near edge)

Mode 0 (PMC w/ IDE) does not appear to work right with ISA interrupts.  The
interrupts from the Winbond chip do not appear at the PMC.

On the PPMC, we assume a 100MHz clock.
on PPMC: (C == closed, or "on")
    SW2:
	C	ROM on PCI bus (DINK32 on mainboard)
	-	Map "B": CHRP
	C	Motorola PPMC
	C	Wait for initialization (peripheral mode)
	-	Program mode: Normal mode
	-	Select normal ROM
	-	33 MHz only
	-	COP only resets local CPU/MPC107
    SW3:
	-C--C	PCI 33, Mem 66, PPC 266
	--	0.5 - 0.9 ns PCI hold time
	C	25 ohm PCI drive strength


Address Map

For this port, we choose the "Address Map B" (CHRP-compatible) for the
system (see SW2, #2, above):

 (Processor View)
0000 0000   0009 FFFF	System Memory
000A 0000   000F FFFF	Compatibility Hole (programmable to go to PCI space
			or system memory--programmed for system memory--cf 5.8)
0010 0000   3FFF FFFF	System memory
4000 0000   7FFF FFFF	Reserved (programmed to give a memory select
			error if accessed--cf 5.7.2)
8000 0000   FCFF FFFF	PCI memory space
FD00 0000   FDFF FFFF	PCI/ISA memory space (see 5.8, CPU_FD_ALIAS_EN)
FE00 0000   FE7F FFFF	PCI/ISA I/O space (Forwarded to PCI address space
			with high byte zeroed, but FE01 0000 and up are
			reserved)
FE80 0000   FEBF FFFF	PCI I/O space (Forwarded to PCI I/O space with high
			byte zeroed)
FEC0 0000   FEDF FFFF	PCI configuration address register (Each word in this
			range is aliased to the PCI CONFIG_ADDR register)
FEE0 0000   FEEF FFFF	PCI configuration data register (Each word in this
			range is aliased to the PCI CONFIG_DATA register)
FEF0 0000   FEFF FFFF	PCI interrupt acknowledge
FF00 0000   FF7F FFFF	32- or 64-bit Flash/ROM space (Can hit either local
			memory or PCI bus -- cf. 5.6)
FF80 0000   FFFF FFFF	8-, 32- or 64-bit Flash/ROM space (Can hit either
			local memory or PCI bus -- cf. 5.6)

This is a host-mode port, so the inbound and output translation windows
are unused.

The Embedded Utilities Memory Block (EUMB) is set to be 1M below the end
of the PCI memory space: FC00 0000, so EUMBBAR is FC00 0000, giving us

Message unit (I2O) base	: FC00 0000	(cf. 10.2, 10.2.3, 10.3)
DMA base		: FC00 1000	(cf. 9.2)
ATU base		: FC00 2000	(cf. 4.3.3)
I2C base		: FC00 3000	(cf. 11.3)
EPIC base		: FC04 0000	(cf. 12.2)



Boot Information

The SandPoint ships with the Motorola DINK32 ROM.  This is a rather
basic ROM with only serial-download (S-Record) capability for
loading the kernel.  Basically, the kernel is loaded to a specified
address and you jump to it.  The ROM takes care of initializing
the MICRs and MCCRs.  There is really no boot information to pass.

It would be nice to have a much more complete ROM interface, allowing
settings for, say, bootp/tftp boot, automatic boot, and persistent
settings (for console rate, auto boot, bootp, etc), and that might
be provided at some point, but that's not available as of this
writing.

So, the kernel is hard-coded to boot w/ 32MB for now.



Interrupt Configuration

The 8240 has the internal EPIC.  For the SandPoint, the EPIC is programmed
in mixed-mode (GCR) with direct interrupts (EICR).  With this configuration,
there are 13 available interrupts:
	4 global timers
	5 direct IRQs
		IRQ0 - PCI Slot #0 INTA#
		IRQ1 - PCI Slot #1 INTA# / shared with WinBond I/O
		IRQ2 - PCI Slot #2 INTA#
		IRQ3 - PCI Slot #3 INTA#
		IRQ4 - On-PPMC 16552 interrupt (Unity X2)
		IRQ4 - pulled down w/ resistor (Unity X4)
	4 internal interrupts
		I2C
		DMA Ch0
		DMA Ch1
		I2O message unit

The SandPoint can run in one of 4 interrupt modes:
  0 - PMC host with IDE (3.3v PCI slots are unavailable)
  1 - PMC host w/o IDE (all PCI slots are available)
  2 - PMC agent, Winbond providing arbitration & interrupt to INTA# on PMC
  3 - Yellowknife mode--just like #2, except drives INTA# on 4th PCI slot

We choose to run in mode 1 as Motorola recommends modes 0 or 1 for
all new development.  Unfortunately, mode 0 does not appear to
work--"ISA" interrupts are lost.  In this mode, with interrupts
routed to PCI slot 3, we have to check for both a Winbond (ISA)
interrupt, and a PCI slot interrupt.  So basically, we have a
two-level interrupt configuration for Winbond interrupts.  The ISA
bus attachment registers an interrupt for PCI slot 3 with its own
interrupt handler.  Drivers for ISA devices on the Winbond will
register interrupts with the ISA interrupt handler.  The sticky
part of this is how to deal with one global interrupt priority.