Old Indy, 64-bit setup

Old Indy, 64-bit setup

[gigo]

  Hello,
Just a silly question. Is there any working 64-bit kernel configuration 
for my r4k 100MHz Indy? From time to time i compile another new kernel for 
64-bit... and see the thing dying. Recently it looked pretty well like 
Indy r4k 100MHz (BROTHER!?!?!) crash shown in 
http://www.linux-mips.org/archives/linux-mips/2007-11/msg00186.html
Everything but the kernel is debian stable: binutils 2.17
gcc 4.1.2 20061115 (prerelease) (Debian 4.1.1-21)...

I would be very happy to give you any more info/do any test, if 
you are interested.

These are taken on 32-bit (usually working) 2.6.23.14 from linux-mips.org.
/proc/cpuinfo:
system type             : SGI Indy
processor               : 0
cpu model               : R4000PC V3.0  FPU V0.0
BogoMIPS                : 45.68
wait instruction        : no
microsecond timers      : yes
tlb_entries             : 48
extra interrupt vector  : no
hardware watchpoint     : yes
ASEs implemented        :
shadow register sets    : 1
VCED exceptions         : 0
VCEI exceptions         : 0

dmesg:
Linux version 2.6.23.14 (root@indyk) (gcc version 4.1.2 20061115 
(prerelease) (Debian 4.1.1-21)) #2 PREEMPT Fri Jan 18 13:00:11 CET 2008
ARCH: SGI-IP22
PROMLIB: ARC firmware Version 1 Revision 10
console [early0] enabled
CPU revision is: 00000430
FPU revision is: 00000500
MC: SGI memory controller Revision 3
MC: Probing memory configuration:
  bank0:  32M @ 08000000
  bank1:  32M @ 0a000000
Determined physical RAM map:
  memory: 04000000 @ 08000000 (usable)
Wasting 1048576 bytes for tracking 32768 unused pages
On node 0 totalpages: 49152
   Normal zone: 384 pages used for memmap
   Normal zone: 0 pages reserved
   Normal zone: 48768 pages, LIFO batch:15
   Movable zone: 0 pages used for memmap
Built 1 zonelists in Zone order.  Total pages: 48768
Kernel command line: root=/dev/sda1 auto
Primary instruction cache 8kB, physically tagged, direct mapped, linesize 16 bytes.
Primary data cache 8kB, direct mapped, linesize 16 bytes.
Synthesized TLB refill handler (21 instructions).
Synthesized TLB load handler fastpath (33 instructions).
Synthesized TLB store handler fastpath (33 instructions).
Synthesized TLB modify handler fastpath (32 instructions).
PID hash table entries: 1024 (order: 10, 4096 bytes)
Calibrating system timer... 48800 [100.0000 MHz CPU]
Using 49.971 MHz high precision timer.
NG1: Revision 3, 8 bitplanes, REX3 revision B, VC2 revision A, xmap9 
revision A, cmap revision C, bt445 revision A
NG1: Screensize 1296x1024
Console: colour SGI Newport 162x64
console handover: boot [early0] -> real [tty0]
Dentry cache hash table entries: 32768 (order: 5, 131072 bytes)
Inode-cache hash table entries: 16384 (order: 4, 65536 bytes)
Memory: 60896k/65536k available (2158k kernel code, 4484k reserved, 325k 
data, 104k init, 0k highmem)
Calibrating delay loop... 45.68 BogoMIPS (lpj=22272)
(...)
Time: MIPS clocksource has been installed.
(...)
io scheduler cfq registered (default)
DS1286 Real Time Clock Driver v1.0
(...)
wd33c93-0: chip=WD33c93B/13 no_sync=0xff no_dma=0 debug_flags=0x00
            setup_args=,,,,,,,,,,
            Version 1.26++ - 10/Feb/2007, Compiled Jan 18 2008 at 06:02:09
scsi0 : SGI WD93
  sending SDTR 010301320c 010301320c sync_xfer=2c
(...)



Regards,

Grzegorz Wieczorek

Re: Old Indy, 64-bit setup

[Thomas Bogendoerfer]

On Tue, Jan 22, 2008 at 10:10:37PM +0100, gigo@poczta.ibb.waw.pl wrote:
> Just a silly question. Is there any working 64-bit kernel configuration 
> for my r4k 100MHz Indy? From time to time i compile another new kernel for 
> 64-bit... and see the thing dying. Recently it looked pretty well like 

your CPU needs a special gcc to avoid triggering 64bit CPU bugs. There
are also some kernel workarounds missing, which are scheduled for 2.6.25.
No idea about the gcc part.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]

Re: Old Indy, 64-bit setup

[Florian Lohoff]

[-- Attachment #1: Type: text/plain, Size: 970 bytes --]

On Tue, Jan 22, 2008 at 10:10:37PM +0100, gigo@poczta.ibb.waw.pl wrote:
>  Hello,
> Just a silly question. Is there any working 64-bit kernel configuration 
> for my r4k 100MHz Indy? From time to time i compile another new kernel for 
> 64-bit... and see the thing dying. Recently it looked pretty well like 
> Indy r4k 100MHz (BROTHER!?!?!) crash shown in 
> http://www.linux-mips.org/archives/linux-mips/2007-11/msg00186.html
> Everything but the kernel is debian stable: binutils 2.17
> gcc 4.1.2 20061115 (prerelease) (Debian 4.1.1-21)...

There are CPU Bug workarounds needed which are not yet complete in
kernel and gcc. I am running my 100Mhz Indy with an 32Bit Kernel built
from the debian source package just by replacing the 64BIT ...

Flo
-- 
Florian Lohoff                  flo@rfc822.org             +49-171-2280134
	Those who would give up a little freedom to get a little 
          security shall soon have neither - Benjamin Franklin

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: Old Indy, 64-bit setup

[Maciej W. Rozycki]

On Tue, 22 Jan 2008, Thomas Bogendoerfer wrote:

> > Just a silly question. Is there any working 64-bit kernel configuration 
> > for my r4k 100MHz Indy? From time to time i compile another new kernel for 
> > 64-bit... and see the thing dying. Recently it looked pretty well like 
> 
> your CPU needs a special gcc to avoid triggering 64bit CPU bugs. There
> are also some kernel workarounds missing, which are scheduled for 2.6.25.
> No idea about the gcc part.

 I have made suitable GCC packages available at:

ftp://ftp3.ds.pg.gda.pl/people/macro/RPMS/

I have just updated the archive and you need at least revision 5 of GCC 
4.1.2 packages available there.  You also need at least revision 2 of 
binutils 2.18 packages to complement your setup.  Pick whichever 
architecture suits you (there are native packages for MIPS configurations 
as well as cross-tools from i386 there) or alternatively grab the 
corresponding source packages from:

ftp://ftp3.ds.pg.gda.pl/people/macro/SRPMS/

and build binaries yourself (that may require some trickery, but you can 
also apply patches manually and use your usual procedure).

 Please note that the relevant kernel changes have only been tested with a 
DECstation, which in particular means little endianness only.  You will 
also have to add this fragment:

select CPU_DADDI_WORKAROUNDS if 64BIT
select CPU_R4000_WORKAROUNDS if 64BIT
select CPU_R4400_WORKAROUNDS if 64BIT

to the right part of the Kconfig file relevant for your machine.  See the 
MACH_DECSTATION section in arch/mips/Kconfig for a reference.  Finally, 
running the kernel from the XPHYS segment is not supported with these 
workarounds enabled, which may worth noting for some platforms, though may 
not necessarily be relevant for yours.

 I plan to update GCC and the associated patches to version 4.2 at some 
point in the future, but I have not decided yet when it is going to happen 
as I am going to stick with 4.1 for a while due to some other work.

 I'll be pleased to get some feedback and good luck!

  Maciej