Re: MVME 5100

Re: MVME 5100

[Matt Porter]

On Wed, Dec 13, 2000 at 10:39:59PM +0100, Torsten Rissel wrote:
>
> Hi Everybody !
>
> I am trying to run Linux on an MVME5100 board. I am using PReP-style booting and
> got the boot process and some of the kernel initialization running. But I am missing any
> knowledge of other PowerPC boards or machines, so that it is relativly hard to find out
> what exactly to change. So I have a question: The MVME5100 is missing all ISA stuff.
> Is the IRQ mapping (first 16 for the 8259 and the other 16 for the MPIC) still valid ?

Might as well just use 16-31 and stick with the MPIC convention.  If
somebody else is using the 5100 with the IPMC761 then they will have
ISA and will thank you many times for not changing things.

> Also I run in a problem: the indirect mechanism for generating PCI config cycles is set
> up correctly, but the board seems to hang when reading from the CONFIG_DATA address !!!???

Make sure you have those Hawk registers covered with a BAT or PTEs.
Notice the areas covered by BATs in the MACH_prep BAT setup case in
arch/ppc/mm/init.c.

--
Matt Porter
MontaVista Software, Inc.
mporter@mvista.com

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

MVME 5100

[Torsten Rissel]

Hi Everybody !

I am trying to run Linux on an MVME5100 board. I am using PReP-style booting and
got the boot process and some of the kernel initialization running. But I am missing any
knowledge of other PowerPC boards or machines, so that it is relativly hard to find out
what exactly to change. So I have a question: The MVME5100 is missing all ISA stuff.
Is the IRQ mapping (first 16 for the 8259 and the other 16 for the MPIC) still valid ?

Also I run in a problem: the indirect mechanism for generating PCI config cycles is set
up correctly, but the board seems to hang when reading from the CONFIG_DATA address !!!???

I would more then appreciate any hints.

Torsten

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

AW: MVME 5100

[Torsten Rissel]

Thanks, but I have already set up a BAT to cover the area from 0xf0000000 - 0xffffffff
because everything is there on the MVME5100 - the serial ports (via the Hawk external
registers) [fef88000+], the MPIC [f3f80000+] and the PHC PCI registers [fe000000+].

The CONFIG_ADDRESS and CONFIG_DATA registers are on fe000cf8 and fe00cfc,
and still writing to the CONFIG_ADDRESS and then reading the CONFIG_DATA sometime
works and sometimes not !!!??

Torsten

-----Ursprungliche Nachricht-----
Von:	Matt Porter [SMTP:mporter@mvista.com]
Gesendet am:	Mittwoch, 13. Dezember 2000 21:41
An:	Torsten Rissel
Cc:	Linuxppc-Embedded (E-Mail)
Betreff:	Re: MVME 5100


On Wed, Dec 13, 2000 at 10:39:59PM +0100, Torsten Rissel wrote:
>
> Hi Everybody !
>
> I am trying to run Linux on an MVME5100 board. I am using PReP-style booting and
> got the boot process and some of the kernel initialization running. But I am missing any
> knowledge of other PowerPC boards or machines, so that it is relativly hard to find out
> what exactly to change. So I have a question: The MVME5100 is missing all ISA stuff.
> Is the IRQ mapping (first 16 for the 8259 and the other 16 for the MPIC) still valid ?

Might as well just use 16-31 and stick with the MPIC convention.  If
somebody else is using the 5100 with the IPMC761 then they will have
ISA and will thank you many times for not changing things.

> Also I run in a problem: the indirect mechanism for generating PCI config cycles is set
> up correctly, but the board seems to hang when reading from the CONFIG_DATA address !!!???

Make sure you have those Hawk registers covered with a BAT or PTEs.
Notice the areas covered by BATs in the MACH_prep BAT setup case in
arch/ppc/mm/init.c.

--
Matt Porter
MontaVista Software, Inc.
mporter@mvista.com

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: MVME 5100

[Matt Porter]

On Thu, Dec 14, 2000 at 04:25:55PM +0100, Torsten Rissel wrote:
> Thanks, but I have already set up a BAT to cover the area from 0xf0000000 - 0xffffffff
> because everything is there on the MVME5100 - the serial ports (via the Hawk external
> registers) [fef88000+], the MPIC [f3f80000+] and the PHC PCI registers [fe000000+].

Right, I have one as well as a similar Hawk-based board that has Linux
on it.

> The CONFIG_ADDRESS and CONFIG_DATA registers are on fe000cf8 and fe00cfc,
> and still writing to the CONFIG_ADDRESS and then reading the CONFIG_DATA sometime
> works and sometimes not !!!??

I can consistently read from and write to config space on my 5100 board
via PPCBUG.  The only thing I can think of is that you have somehow
locked up the Hawk bridge.  I've managaged this on other host bridges
only by getting a PIB locked up and squatting on the bus.

What kernel version are you using as a basis for your port?  Are you
basing the port off the existing prep_*.c support?

--
Matt Porter
MontaVista Software, Inc.
mporter@mvista.com

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: MVME 5100

[Gabriel Paubert]

On Fri, 15 Dec 2000, Matt Porter wrote:

>
> On Thu, Dec 14, 2000 at 04:25:55PM +0100, Torsten Rissel wrote:
> > Thanks, but I have already set up a BAT to cover the area from 0xf0000000 - 0xffffffff
> > because everything is there on the MVME5100 - the serial ports (via the Hawk external
> > registers) [fef88000+], the MPIC [f3f80000+] and the PHC PCI registers [fe000000+].
>
> Right, I have one as well as a similar Hawk-based board that has Linux
> on it.
>
> > The CONFIG_ADDRESS and CONFIG_DATA registers are on fe000cf8 and fe00cfc,
> > and still writing to the CONFIG_ADDRESS and then reading the CONFIG_DATA sometime
> > works and sometimes not !!!??
>
> I can consistently read from and write to config space on my 5100 board
> via PPCBUG.  The only thing I can think of is that you have somehow
> locked up the Hawk bridge.  I've managaged this on other host bridges
> only by getting a PIB locked up and squatting on the bus.

On the MVME2400 (and older Raven/Falcon based boards) the config space by
default is at 0x80000cf8 and 0x8000cfc. Does it mean that on the MVME5100,
the Hawk is reprogrammed by PPCBUG to look like a CHRP system ?

	Regards,
	Gabriel.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: MVME 5100

[Matt Porter]

On Mon, Dec 18, 2000 at 04:34:57PM +0100, Gabriel Paubert wrote:
>
>
> On Fri, 15 Dec 2000, Matt Porter wrote:
>
> >
> > On Thu, Dec 14, 2000 at 04:25:55PM +0100, Torsten Rissel wrote:
> > > Thanks, but I have already set up a BAT to cover the area from 0xf0000000 - 0xffffffff
> > > because everything is there on the MVME5100 - the serial ports (via the Hawk external
> > > registers) [fef88000+], the MPIC [f3f80000+] and the PHC PCI registers [fe000000+].
> >
> > Right, I have one as well as a similar Hawk-based board that has Linux
> > on it.
> >
> > > The CONFIG_ADDRESS and CONFIG_DATA registers are on fe000cf8 and fe00cfc,
> > > and still writing to the CONFIG_ADDRESS and then reading the CONFIG_DATA sometime
> > > works and sometimes not !!!??
> >
> > I can consistently read from and write to config space on my 5100 board
> > via PPCBUG.  The only thing I can think of is that you have somehow
> > locked up the Hawk bridge.  I've managaged this on other host bridges
> > only by getting a PIB locked up and squatting on the bus.
>
> On the MVME2400 (and older Raven/Falcon based boards) the config space by
> default is at 0x80000cf8 and 0x8000cfc. Does it mean that on the MVME5100,
> the Hawk is reprogrammed by PPCBUG to look like a CHRP system ?

Yes...and Yes. :)  The only thing the newer MCG boards really do
(correctly) that looks like a PReP system is the disk booting
mechanism.

BTW, the MVME2100 (8240/Universe II board) is put in map B for a CHRP
map as well.

--
Matt Porter
MontaVista Software, Inc.
mporter@mvista.com

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: MVME 5100

[Gabriel Paubert]

On Mon, 18 Dec 2000, Matt Porter wrote:

> > On the MVME2400 (and older Raven/Falcon based boards) the config space by
> > default is at 0x80000cf8 and 0x8000cfc. Does it mean that on the MVME5100,
> > the Hawk is reprogrammed by PPCBUG to look like a CHRP system ?
>
> Yes...and Yes. :)  The only thing the newer MCG boards really do
> (correctly) that looks like a PReP system is the disk booting
> mechanism.
>
> BTW, the MVME2100 (8240/Universe II board) is put in map B for a CHRP
> map as well.

Ok, fine. It means that  my code to reprogram the Raven/Hawk on 2[467]xx
just makes things more uniform. I still don't like the fact that the DMA
is offset, though (unless it's a residual data bug, and map B on 106/107
does not have this DMA offset).

	Regards,
	Gabriel.


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/