MIPS64

MIPS64

[Ralf Baechle]

Hi,

as those who are tracking the CVS archive or the commit mailing list
probably already have seen I've got started to work on a 64-bit kernel.
I'm also using the chance to do a major overhaul of various code which
over the years had turned into a major uglyness.

In particular the machine abstraction layer needs some overhaul.
Basically the attempt of doing generic kernels for MIPS is dead by now.
Given the number of different firmware implementations and memory layouts
makes it impossible to build a single generic kernel image that fits
more than a few machines.  In fact so far the only combo that was working
at all was Magnum 4000 / PICA / RM200C.  So I'm wiping out all that code
which makes things look significantly nicer.

As of now the MIPS64 code is still in it's very early stages; it doesn't
even compile.  However I'd like to invite other people to alreasy start
working on porting the machine specific bits to MIPS64.  I myself will
initially only implement support for the SGI IP22.  When this one is
running flawless I'll go on with IP27 and SMP support.

Even though I've invested a horrible amount of time int tracking down
the bugs in current binutils they're still in a rather sad state for
64-bit stuff.  Volunteers _urgently_ wanted.  Note that these problems
will also affect other operating system, so it doesn't necessarily have
to be a Linux volunteer.

Cheers,

  Ralf

Re: MIPS64

[Ralf Baechle]

On Sun, Aug 22, 1999 at 04:58:33PM -0400, Bradley D. LaRonde wrote:

> Good news.  I think this is a good move, especially if a flood of new ports
> to new device starts pouring in from our project (LinuxCE).  There are
> already three new platoforms and two new processors on the drawing board
> now.

Note that it's not compeltly impossible, just unattractive to make a
generic kernel.  So if necessary we can do it.

  Ralf

Re: MIPS64

[Ralf Baechle]

On Sun, Aug 22, 1999 at 06:12:25PM -0400, Bradley D. LaRonde wrote:

> > > Good news.  I think this is a good move, especially if a flood of new
> ports
> > > to new device starts pouring in from our project (LinuxCE).  There are
> > > already three new platoforms and two new processors on the drawing board
> > > now.
> >
> > Note that it's not compeltly impossible, just unattractive to make a
> > generic kernel.  So if necessary we can do it.
> 
> I'm also thinking that platform-specifc kernels will be smaller and
> potentially faster (less indirection), which is especially important on
> platforms like the Clio and Cassiopea.

For many systems we cannot drop all indirections.  For example the Indy
shipped with a large number of CPU options, each of which we have an
optimized routine for.

  Ralf

Re: MIPS64

[Andreas Jaeger]

>>>>> Ralf Baechle writes:

 > Hi,
 > as those who are tracking the CVS archive or the commit mailing list
 > probably already have seen I've got started to work on a 64-bit kernel.
 > I'm also using the chance to do a major overhaul of various code which
 > over the years had turned into a major uglyness.
Could anybody provide me with infos about the commit mailing list?  I
couldn't find the information myself:-(.

Thanks,
Andreas
-- 
 Andreas Jaeger   aj@arthur.rhein-neckar.de    jaeger@informatik.uni-kl.de
  for pgp-key finger ajaeger@aixd1.rhrk.uni-kl.de

Re: MIPS64

[Ralf Baechle]

On Mon, Aug 23, 1999 at 04:28:54PM +0200, Andreas Jaeger wrote:

>  > as those who are tracking the CVS archive or the commit mailing list
>  > probably already have seen I've got started to work on a 64-bit kernel.
>  > I'm also using the chance to do a major overhaul of various code which
>  > over the years had turned into a major uglyness.
> Could anybody provide me with infos about the commit mailing list?  I
> couldn't find the information myself:-(.

Yep, because it's not documented.  The mailing list is a plain
sendmail alias.  But I think you can also subscribe to it by

echo "subscribe linux-progress" | mail majordomo@engr.sgi.com

  Ralf

Re: MIPS64

[Ralf Baechle]

On Thu, Oct 07, 1999 at 02:20:22PM +0200, Ludek Finstrle wrote:

> Oh, where is online? I can't see it anywhere :o(

oss.sgi.com:/pub/pub/linux/mips/src/binutils-19990825.tar.bz2 are the
sources for the binutils which I'm using.  Fate doesn't like me, I
lost connection to oss when I tried to upload the egcs 1.1.2 patch
for MIPS64.  I'll put it into that directory tomorrow.

  Ralf

Re: MIPS64

[Ludek Finstrle]

> > > Oh, where is online? I can't see it anywhere :o(
> > 
> > oss.sgi.com:/pub/pub/linux/mips/src/binutils-19990825.tar.bz2 are the
> > sources for the binutils which I'm using.  Fate doesn't like me, I
> > lost connection to oss when I tried to upload the egcs 1.1.2 patch
> > for MIPS64.  I'll put it into that directory tomorrow.
> 
> Ok, things are now online.  While I was at it I also have copied all
> the stuff from the old ftp.linux.sgi.com over to oss.sgi.com.  The
> crosscompiler source & patches are now in
> oss.sgi.com:/pub/linux/mips/src/mips64/.

Hello,

  we have some problems with code from cross-compiler for MIPS64.
We download sources from cygnus CVS and applied your patch. We had
some problems with compile egcs. We fix it but it generate bad
code => we fix it bad :o((

  Could you give us source from which you compiled your egcs cross-
compiler.

Thanks a lot

Luf

mips64

[Michael Dosser]

Hi,

I'm successfully running Debian/GNU Linux on a SGI Indy R4600PC@100Mhz
for over a year now. I'm very happy with the stability of Linux on that
machine. But since the machine is relatively slow (currently 30-35 
shell user continuosly connected), I bought an Indigo2 R4400SC@250Mhz.

I thought of putting a mips64 kernel on the new machine: Got the rpm's
from ftp.linux-mips.org, converted them with alien to debs and installed 
them on my quad xeon Debian box - checked out the linux source and 
started compiling:

# cvs -d :pserver:cvs@ftp.linux-mips.org:/home/cvs co -r linux_2_4 linux
# cd linux
# make ARCH=mips64 dep
# make ARCH=mips64 clean
# make ARCH=mips64 all

Error message with gcc version egcs-2.91.66 19990314 (egcs-1.1.2
release)

[...]
make[2]: Entering directory `/usr/local/src/mips/linux/arch/mips/math-emu'
mips64-linux-gcc -D__KERNEL__ -I/usr/local/src/mips/linux/include -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -I /usr/local/src/mips/linux/include/asm/gcc -mabi=64 -G 0 -mno-abicalls -fno-pic -Wa,--trap -pipe -mcpu=r4600 -mips3 -Wa,-32 -Wa,-mgp64   -nostdinc -iwithprefix include -DKBUILD_BASENAME=cp1emu  -c -o cp1emu.o cp1emu.c
cp1emu.c: In function `fpu_emulator_cop1Handler':
cp1emu.c:1328: internal error--unrecognizable insn:
(insn 310 33 25 (set (reg:SI 159)
        (reg/v:DI 87)) -1 (insn_list:REG_DEP_ANTI 28 (insn_list 33 (nil)))
    (nil))
../../gcc/toplev.c:1367: Internal compiler error in function fatal_insn
make[2]: *** [cp1emu.o] Error 1
make[2]: Leaving directory `/usr/local/src/mips/linux/arch/mips/math-emu'
make[1]: *** [first_rule] Error 2
make[1]: Leaving directory `/usr/local/src/mips/linux/arch/mips/math-emu'
make: *** [_dir_arch/mips/math-emu] Error 2
# 

Error with gcc version 2.95.4 20010319 (prerelease):

[...]
mips64-linux-ld --oformat elf32-tradbigmips   -r -o kernel.o sched.o
dma.o fork.o exec_domain.o panic.o printk.o module.o exit.o itimer.o
info.o time.o softirq.o resource.o sysctl.o acct.o capability.o ptrace.o
timer.o user.o signal.o sys.o kmod.o context.o ksyms.o
mips64-linux-ld: target elf32-tradbigmips not found
make[2]: *** [kernel.o] Error 1
make[2]: Leaving directory `/usr/local/src/mips/linux/kernel'
make[1]: *** [first_rule] Error 2
make[1]: Leaving directory `/usr/local/src/mips/linux/kernel'
make: *** [_dir_kernel] Error 2
#

Ok, the latter seems to be related to objdump, right?
mips64-linux-objdump: supported targets: elf32-bigmips elf32-littlemips
elf64-bigmips elf64-littlemips ecoff-bigmips ecoff-littlemips
elf64-little elf64-big elf32-little elf32-big srec symbolsrec tekhex
binary ihex

The package on linux-mips.org seems not to be including
elf32-tradbigmips ...

Can somebody help me with this? Btw: same errors with co -r linux_2_4_21
...

Thank you,mic

-- 
> Please specifically define where data goes that is sent to /dev/null
[...]
Answer 2.  All the data goes into another dimension, and comes out of
/dev/random.            Stephen Montgomery-Smith on freebsd-hackers

Re: mips64

[Thiemo Seufer]

Michael Dosser wrote:
> Hi,
> 
> I'm successfully running Debian/GNU Linux on a SGI Indy R4600PC@100Mhz
> for over a year now. I'm very happy with the stability of Linux on that
> machine. But since the machine is relatively slow (currently 30-35 
> shell user continuosly connected), I bought an Indigo2 R4400SC@250Mhz.
> 
> I thought of putting a mips64 kernel on the new machine: Got the rpm's
> from ftp.linux-mips.org, converted them with alien to debs and installed 
> them on my quad xeon Debian box - checked out the linux source and 
> started compiling:

The 64bit IP22 Kernel was broken for quite some time, I don't know if
this changed in the meanwhile.

[snip]
> Error message with gcc version egcs-2.91.66 19990314 (egcs-1.1.2
> release)
> 
> [...]
> make[2]: Entering directory `/usr/local/src/mips/linux/arch/mips/math-emu'
> mips64-linux-gcc -D__KERNEL__ -I/usr/local/src/mips/linux/include -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -I /usr/local/src/mips/linux/include/asm/gcc -mabi=64 -G 0 -mno-abicalls -fno-pic -Wa,--trap -pipe -mcpu=r4600 -mips3 -Wa,-32 -Wa,-mgp64   -nostdinc -iwithprefix include -DKBUILD_BASENAME=cp1emu  -c -o cp1emu.o cp1emu.c
> cp1emu.c: In function `fpu_emulator_cop1Handler':
> cp1emu.c:1328: internal error--unrecognizable insn:
> (insn 310 33 25 (set (reg:SI 159)
>         (reg/v:DI 87)) -1 (insn_list:REG_DEP_ANTI 28 (insn_list 33 (nil)))
>     (nil))
> ../../gcc/toplev.c:1367: Internal compiler error in function fatal_insn

egcs is known to be broken WRT, and horribly outdated anyway.

[snip]
> Error with gcc version 2.95.4 20010319 (prerelease):
> 
> [...]
> mips64-linux-ld --oformat elf32-tradbigmips   -r -o kernel.o sched.o
> dma.o fork.o exec_domain.o panic.o printk.o module.o exit.o itimer.o
> info.o time.o softirq.o resource.o sysctl.o acct.o capability.o ptrace.o
> timer.o user.o signal.o sys.o kmod.o context.o ksyms.o
> mips64-linux-ld: target elf32-tradbigmips not found
> make[2]: *** [kernel.o] Error 1
> make[2]: Leaving directory `/usr/local/src/mips/linux/kernel'
> make[1]: *** [first_rule] Error 2
> make[1]: Leaving directory `/usr/local/src/mips/linux/kernel'
> make: *** [_dir_kernel] Error 2
> #
> 
> Ok, the latter seems to be related to objdump, right?

No, this ld is too old to handle elf32-tradbigmips.
You'll need a more up to date toolchain.


Thiemo

Re: mips64

[Ladislav Michl]

On Wed, Aug 20, 2003 at 12:15:09PM +0200, Thiemo Seufer wrote:
> Michael Dosser wrote:
> > Hi,
> > 
> > I'm successfully running Debian/GNU Linux on a SGI Indy R4600PC@100Mhz
> > for over a year now. I'm very happy with the stability of Linux on that
> > machine. But since the machine is relatively slow (currently 30-35 
> > shell user continuosly connected), I bought an Indigo2 R4400SC@250Mhz.
> > 
> > I thought of putting a mips64 kernel on the new machine: Got the rpm's
> > from ftp.linux-mips.org, converted them with alien to debs and installed 
> > them on my quad xeon Debian box - checked out the linux source and 
> > started compiling:
> 
> The 64bit IP22 Kernel was broken for quite some time, I don't know if
> this changed in the meanwhile.

It works now.

> [snip]
> > Error message with gcc version egcs-2.91.66 19990314 (egcs-1.1.2
> > release)
> > 
> > [...]
> > make[2]: Entering directory `/usr/local/src/mips/linux/arch/mips/math-emu'
> > mips64-linux-gcc -D__KERNEL__ -I/usr/local/src/mips/linux/include -Wall -Wstrict-prototypes -Wno-trigraphs -O2 -fno-strict-aliasing -fno-common -fomit-frame-pointer -I /usr/local/src/mips/linux/include/asm/gcc -mabi=64 -G 0 -mno-abicalls -fno-pic -Wa,--trap -pipe -mcpu=r4600 -mips3 -Wa,-32 -Wa,-mgp64   -nostdinc -iwithprefix include -DKBUILD_BASENAME=cp1emu  -c -o cp1emu.o cp1emu.c
> > cp1emu.c: In function `fpu_emulator_cop1Handler':
> > cp1emu.c:1328: internal error--unrecognizable insn:
> > (insn 310 33 25 (set (reg:SI 159)
> >         (reg/v:DI 87)) -1 (insn_list:REG_DEP_ANTI 28 (insn_list 33 (nil)))
> >     (nil))
> > ../../gcc/toplev.c:1367: Internal compiler error in function fatal_insn
> 
> egcs is known to be broken WRT, and horribly outdated anyway.
> 
> [snip]
> > Error with gcc version 2.95.4 20010319 (prerelease):
> > 
> > [...]
> > mips64-linux-ld --oformat elf32-tradbigmips   -r -o kernel.o sched.o
> > dma.o fork.o exec_domain.o panic.o printk.o module.o exit.o itimer.o
> > info.o time.o softirq.o resource.o sysctl.o acct.o capability.o ptrace.o
> > timer.o user.o signal.o sys.o kmod.o context.o ksyms.o
> > mips64-linux-ld: target elf32-tradbigmips not found
> > make[2]: *** [kernel.o] Error 1
> > make[2]: Leaving directory `/usr/local/src/mips/linux/kernel'
> > make[1]: *** [first_rule] Error 2
> > make[1]: Leaving directory `/usr/local/src/mips/linux/kernel'
> > make: *** [_dir_kernel] Error 2
> > #
> > 
> > Ok, the latter seems to be related to objdump, right?
> 
> No, this ld is too old to handle elf32-tradbigmips.
> You'll need a more up to date toolchain.

gcc-2.95.4 and binutils-2.14 works ok for me.

ladis

Re: mips64

[Michael Dosser]

Hi again,

* On 2003-08-20 12:15 <ica2_ts@csv.ica.uni-stuttgart.de> wrote:

> egcs is known to be broken WRT, and horribly outdated anyway.

That's right. But I thought better trying out first before writing an
email.

> No, this ld is too old to handle elf32-tradbigmips.
> You'll need a more up to date toolchain.

I could slap myself for this: I fetched binutils 2.9.5 instead of 2.13.1
while thinking 9 > 1 and not reading 13. *hmpf* Sorry for this. After
installing binutils-2.13.1 the compile went fine.

# ls -l arch/mips64/boot
total 2256
drwxr-xr-x    2 root     root         4096 Aug 20 11:13 CVS
-rw-r--r--    1 root     root          922 Jan 10  2003 Makefile
-rwxr-xr-x    1 root     root         7965 Aug 20 13:44 addinitrd
-rwxr-xr-x    1 root     root        16138 Aug 20 13:44 elf2ecoff
-rwxr-xr-x    1 root     root      2269392 Aug 20 13:44 vmlinux.ecoff
# file arch/mips64/boot/vmlinux.ecoff
vmlinux.ecoff: MIPSEB ECOFF executable (impure) stripped - version 0.200
#

That's it I suppose :) I'm looking forward to test this kernel ...

Thanks for your help.

Ciao,mic

-- 
while !asleep {sheep ++};