64-bit on Cobalt?

64-bit on Cobalt?

[John Tobey]

Hi folks,

I keep having this fantasy that I will one day make my old Cobalt RaQ
run a Debian system including 64-bit apps.  I've got a copy of _See
MIPS Run_ (wonderful book) and a working system based on Cobalt's
hacked-up kernel 2.0.34 and Red Hat 5.x.  I am comfortable using
cross-gcc and messing with glibc.

It seems that I have all the information required to do it all myself,
but perhaps I might save a few years of work (or rid myself of this
delusion) by consulting with you all first.

The CPU is a QED RM5231 (CONFIG_NEVADA) 150MHz.  May I assume that
nobody has run a 64-bit kernel on this thing?  The RaQ has no video
card but a serial console, PCI, IDE, Ethernet, and special LEDs, panel
buttons, and LCD display.  If I can get a 64-bit kernel to boot and
prove its existence through any of these devices, I will be drunk with
power.

The reason I want 64 bits is that I (a) want a challenge, (b) plan to
write an application that uses a sparse address space (40 bits is
better than 31), (c) plan to outlive the 31-bit time_t, and (d) am
p.o.ed at having bought the thing based on misleading advertising that
mentioned a 64-bit processor but not the 32-bit OS.

Big/little endian macht nichts.  I guess big will be easier, and I'm
not concerned with running any existing 32-bit binaries.

I imagine that I would start by grafting Cobalt's peripheral support
code from arch/mips/cobalt (now defunct) and include/asm-mips/cobalt.h
into the mips64 tree from cvs@oss.sgi.com:/cvs/linux.

I will appreciate your advice.

-John

-- 
John Tobey, late nite hacker <jtobey@john-edwin-tobey.org>
\\\                                                               ///
]]]             With enough bugs, all eyes are shallow.           [[[
///                                                               \\\

Re: 64-bit on Cobalt?

[Ralf Baechle]

On Sun, Apr 08, 2001 at 06:42:41PM -0400, John Tobey wrote:

> The CPU is a QED RM5231 (CONFIG_NEVADA) 150MHz.  May I assume that
> nobody has run a 64-bit kernel on this thing?  The RaQ has no video
> card but a serial console, PCI, IDE, Ethernet, and special LEDs, panel
> buttons, and LCD display.  If I can get a 64-bit kernel to boot and
> prove its existence through any of these devices, I will be drunk with
> power.

So far the only supported machine by the mips64 kernel is the SGI Origin
200 / 2000 series.

> The reason I want 64 bits is that I (a) want a challenge, (b) plan to
> write an application that uses a sparse address space (40 bits is
> better than 31), (c) plan to outlive the 31-bit time_t, and (d) am
> p.o.ed at having bought the thing based on misleading advertising that
> mentioned a 64-bit processor but not the 32-bit OS.

> Big/little endian macht nichts.  I guess big will be easier, and I'm
> not concerned with running any existing 32-bit binaries.

Go for little endian because the firmware is little endian; supporting
``other-endian'' for userspace would be unecessary extra pain.  We already
have suport for 32-bit binaries in the 64-bit kernel; in fact ALL
software we run on 64-bit kernels is 32-bit.

32-bit wasn't only the easy thing to do - it's also the more efficient
thing for most software which doesn't need 64-bit registers or 64-bit
address space.  For a system with a dog slow 32-bit memory bus such as
the Qube 64-bit kernels would mean a dramatic slowdown.

I admit it's interesting though, mostly for engineering reasons, not
as a platform.

> I imagine that I would start by grafting Cobalt's peripheral support
> code from arch/mips/cobalt (now defunct) and include/asm-mips/cobalt.h
> into the mips64 tree from cvs@oss.sgi.com:/cvs/linux.

Somebody else was already working on upgrading the Cobalt kernel to 2.4.

  Ralf

Re: 64-bit on Cobalt?

[Carsten Langgaard]

Ralf Baechle wrote:

> On Sun, Apr 08, 2001 at 06:42:41PM -0400, John Tobey wrote:
>
> > The CPU is a QED RM5231 (CONFIG_NEVADA) 150MHz.  May I assume that
> > nobody has run a 64-bit kernel on this thing?  The RaQ has no video
> > card but a serial console, PCI, IDE, Ethernet, and special LEDs, panel
> > buttons, and LCD display.  If I can get a 64-bit kernel to boot and
> > prove its existence through any of these devices, I will be drunk with
> > power.
>
> So far the only supported machine by the mips64 kernel is the SGI Origin
> 200 / 2000 series.

I have a QED RM5261 running a bigendian 64-bit kernel on our Malta board (it
got a serial console, PCI, IDE, Ethernet).
I haven't send the patch to Ralf yet, but you can get a snapshot at our
kernel sources at
ftp://ftp.mips.com/pub/linux/mips/kernel/2.4/src/linux-2.4.1.mips-src-01.00.tar.gz

Hope it helps you.

>
> > The reason I want 64 bits is that I (a) want a challenge, (b) plan to
> > write an application that uses a sparse address space (40 bits is
> > better than 31), (c) plan to outlive the 31-bit time_t, and (d) am
> > p.o.ed at having bought the thing based on misleading advertising that
> > mentioned a 64-bit processor but not the 32-bit OS.
>
> > Big/little endian macht nichts.  I guess big will be easier, and I'm
> > not concerned with running any existing 32-bit binaries.
>
> Go for little endian because the firmware is little endian; supporting
> ``other-endian'' for userspace would be unecessary extra pain.  We already
> have suport for 32-bit binaries in the 64-bit kernel; in fact ALL
> software we run on 64-bit kernels is 32-bit.
>
> 32-bit wasn't only the easy thing to do - it's also the more efficient
> thing for most software which doesn't need 64-bit registers or 64-bit
> address space.  For a system with a dog slow 32-bit memory bus such as
> the Qube 64-bit kernels would mean a dramatic slowdown.
>
> I admit it's interesting though, mostly for engineering reasons, not
> as a platform.
>
> > I imagine that I would start by grafting Cobalt's peripheral support
> > code from arch/mips/cobalt (now defunct) and include/asm-mips/cobalt.h
> > into the mips64 tree from cvs@oss.sgi.com:/cvs/linux.
>
> Somebody else was already working on upgrading the Cobalt kernel to 2.4.
>
>   Ralf

--
_    _ ____  ___   Carsten Langgaard   Mailto:carstenl@mips.com
|\  /|||___)(___   MIPS Denmark        Direct: +45 4486 5527
| \/ |||    ____)  Lautrupvang 4B      Switch: +45 4486 5555
  TECHNOLOGIES     2750 Ballerup       Fax...: +45 4486 5556
                   Denmark             http://www.mips.com

Re: 64-bit on Cobalt?

[John Tobey]

On Mon, Apr 09, 2001 at 11:37:59AM +0200, Carsten Langgaard wrote:
> 
> I have a QED RM5261 running a bigendian 64-bit kernel on our Malta board (it
> got a serial console, PCI, IDE, Ethernet).
> I haven't send the patch to Ralf yet, but you can get a snapshot at our
> kernel sources at
> ftp://ftp.mips.com/pub/linux/mips/kernel/2.4/src/linux-2.4.1.mips-src-01.00.tar.gz
> 
> Hope it helps you.

Wow, thanks!  It sure looks like it will.

-- 
John Tobey, late nite hacker <jtobey@john-edwin-tobey.org>
\\\                                                               ///
]]]             With enough bugs, all eyes are shallow.           [[[
///                                                               \\\

Re: 64-bit on Cobalt?

[John Tobey]

On Mon, Apr 09, 2001 at 03:54:53AM +0200, Ralf Baechle wrote:
> 
> I admit it's interesting though, mostly for engineering reasons, not
> as a platform.
> 
> > I imagine that I would start by grafting Cobalt's peripheral support
> > code from arch/mips/cobalt (now defunct) and include/asm-mips/cobalt.h
> > into the mips64 tree from cvs@oss.sgi.com:/cvs/linux.
> 
> Somebody else was already working on upgrading the Cobalt kernel to 2.4.

Hey, thanks for the advice!  If anyone's interested, here's a mailing
list for porting recent kernels to Cobalt/MIPS:
http://devel.alal.com/mailman/listinfo/cobalt-22

Best.
-John

64-bit on Origin (was: 64-bit on Cobalt?)

[Lukas Hejtmanek]

On Mon, Apr 09, 2001 at 03:54:53AM +0200, Ralf Baechle wrote:
> So far the only supported machine by the mips64 kernel is the SGI Origin
> 200 / 2000 series.

So it is possible to run 64-bit application on Origin 200?

What's the state of user-land 64-bit applications for that?
(gcc, binutils, glibc)

We have Origin 200 at university and my diploma thesis could be to run true
64-bit linux on it. Anyone interested?

-- 
Lukáš Hejtmánek

Re: 64-bit on Origin (was: 64-bit on Cobalt?)

[Lukas Hejtmanek]

On Fri, Apr 13, 2001 at 01:25:10AM +0200, Ralf Baechle wrote:
> > So it is possible to run 64-bit application on Origin 200?
> 
> In theory yes - if you have any ...

I ment something like this:
file mipstest.o
mipstest.o: ELF 64-bit MSB mips-3 relocatable, MIPS R3000_BE, version 1, not
stripped

> gcc - should work.  Binutils - major brain surgery required.  glibc -
> 64-bit support practically non-existant.

So as and/or ld generates bad code?

Does anyone already work on it?

And if binutils are broken that means kernel is 32-bit as well?


Anyway 32-bit applications should run pretty fine?

-- 
Lukáš Hejtmánek

Re: 64-bit on Origin (was: 64-bit on Cobalt?)

[Ralf Baechle]

On Fri, Apr 13, 2001 at 12:06:12AM +0200, Lukas Hejtmanek wrote:

> So it is possible to run 64-bit application on Origin 200?

In theory yes - if you have any ...

> What's the state of user-land 64-bit applications for that?
> (gcc, binutils, glibc)

gcc - should work.  Binutils - major brain surgery required.  glibc -
64-bit support practically non-existant.

> We have Origin 200 at university and my diploma thesis could be to run true
> 64-bit linux on it. Anyone interested?

  Ralf

Re: 64-bit on Origin (was: 64-bit on Cobalt?)

[Ralf Baechle]

On Fri, Apr 13, 2001 at 12:50:25AM +0200, Lukas Hejtmanek wrote:

> I ment something like this:
> file mipstest.o
> mipstest.o: ELF 64-bit MSB mips-3 relocatable, MIPS R3000_BE, version 1, not
> stripped
> 
> > gcc - should work.  Binutils - major brain surgery required.  glibc -
> > 64-bit support practically non-existant.
> 
> So as and/or ld generates bad code?

Both know fairly little about 64-bit ELF atm.

> Does anyone already work on it?

Thiemo Seufer <ica2_ts@csv.ica.uni-stuttgart.de> is working on it; it's
quite an effort to get it right so if you want to work on this you should
probably team up with him.

> And if binutils are broken that means kernel is 32-bit as well?

Yes but that's more a feature than a bug.  All intra-kernel references are
now only 32-bit therefore we've got more compact, faster code as result.

The boot file is a 64-bit ELF kernel, though.  That's because the ARC
firmware of the Origin will accept nothing else so we cheat by converting
the 32-bit vmlinux into a 64-bit file using objcopy.

> Anyway 32-bit applications should run pretty fine?

Most of them.  Now and then we find bugs in the binary compatibility code
but these are usually easy to squash.

   Ralf