Same boot problem on a 9000/720/50 HP machine as on 712/60 of Niskanen

On my 9000/720 (I have it since one week) I have the same problem booting the CD 0.5
or
a selfbuild kernel. The kernel boots, but on calling the init a trap is called with the text:

handle_interruption() pid=1 command='init'

YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
PSW: 00000000000001001111111100001011

r0-3     00000000 000c0f38 00033a1b 000c0f3c
r4-7     000c0f38 ffffffff 00000001 0002a1d8
r8-11    c02ac640 00000000 00000000 c0145f30
r12-15   00000000 00000000 00000000 00000000
r16-19   c027b60c 00000000 c02ac000

and so on, running in a loop.

The code is in kernel/traps.c

Further I tried to build a statically linked bash-2.03, but I had problems setting up the build process to produce a HP PARISC user program. What I have to do to setup or prepare default source trees for cross building (the compiler runs....)

Cheers

Christoph Plattner

-------------------------------------------------------------------------
private:	[email protected]
company:	[email protected]

Christoph Plattner writes...

Yes, several people(myself included) saw this. IIRC a newer kernel fixes it.

Which compiler are you using? You can grab a prebuilt hppa cross-compiler (that runs on i386-linux) at,

ftp://puffin.external.hp.com/pub/parisc/binaries/LinuxX86/xc-latest.tar.gz

bash is particularly hard to build. I was able to get the debian package to cross compile using the dpkg-cross tool and by making a few changes to the source tree. I still need to test the binary, if it works I'll post it.

Matt Taggart
[email protected]

Matt Taggart writes...

[snip init looping]

Ok, I got bash to build static against the glibc in the xc and I setup an nfsroot with bash as my init. I am using a new kernel built with the same xc. Here's what I get when booting,

-----------------------------------------------------------------------------
.
.
.
IP-Config: Got BOOTP answer from 15.1.51.72, my address is 15.1.51.73
NET4: Unix domain sockets 1.0/SMP for Linux NET4.0.
kmem_create: Forcing size word alignment - nfs_fh
Looking up port of RPC 100003/2 on 15.1.51.72
Looking up port of RPC 100005/2 on 15.1.51.72
VFS: Mounted root (nfs filesystem) readonly.

do_page_fault() pid=1 command='init' address=0xfffffff8 vm_start = 0xbb000000, vm_end = 0xbb002000

YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
PSW: 00000000000001001111111100001011

r0-3     00000000 001071d0 00088153 fffffff8
r4-7     bb0002c8 00000023 00000000 0000003a
r8-11    00001368 00000000 000d6884 c01456a4
r12-15   00000000 00000000 00000000 00000000
r16-19   c027b60c c02ac000 00000000 00000001
r20-23   000e3848 000e384c 01010101 80808080
r24-27   000e3828 00000023 00000000 000fd9d0
r28-31   00000022 0000000a bb000480 0008d443
sr0-3    00000000 00000000 00000000 00000001
sr4-7    00000001 00000001 00000001 00000001

IASQ: 00000001 00000001 IAOQ: 00068697 0006869b
IIR: 0c601084 ISR: 00000001 IOR: fffffff8
ORIG_R28: 00000003
-----------------------------------------------------------------------------

and it keeps looping with that same error.

The bash I'm using is at

ftp://puffin.external.hp.com/pub/parisc/debug/bash-20001226.tar.gz

The xc I'm using is at

ftp://puffin.external.hp.com/pub/parisc/debug/xc-20001222.tar.gz

I think the kernel and libc have different world views. I can provide different bits on request.

Any ideas?

Matt Taggart
[email protected]

Matt Taggart writes...

I also just tried booting an old nfsroot, then trying to run the static bash. Here's what I get,

-----------------------------------------------------------------------------

geordi:/tmp/bash/bin# ./bash

do_page_fault() pid=117 command='bash' address=0xfffffff8 vm_start = 0xbb000000, vm_end = 0xbb001000

YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
PSW: 00000000000001001111111100001011

r0-3     00000000 001071d0 00088153 fffffff8
r4-7     bb000408 00000023 00000000 0000003a
r8-11    00001368 0008b810 000d6884 00094cf0
r12-15   00000000 00000000 00094cb0 00000000
r16-19   00000000 00000001 0000b71b 00000001
r20-23   000e3848 000e384c 01010101 80808080
r24-27   000e3828 00000023 00000000 000fd9d0
r28-31   00000022 000ec839 bb0005c0 0008d443
sr0-3    00000000 00000000 00000000 00000003
sr4-7    00000003 00000003 00000003 00000003

IASQ: 00000003 00000003 IAOQ: 00068697 0006869b
IIR: 0c601084 ISR: 00000003 IOR: fffffff8
ORIG_R28: 00000003
Segmentation fault
-----------------------------------------------------------------------------

Just to see how bad the problem was I built helloworld and it runs ok. I'll try to narrow it down more than that :)

Matt Taggart
[email protected]

Matt Taggart writes...

[bash problems snipped]

It seems to just be the xcompiled static bash I built. I xcompiled sash and it worked ok. I setup an nfsroot with a bunch of stuff I had cross-compiled and except for bash and a couple other things, it's working normally. I will try compiling bash native and see if that helps.

Matt Taggart
[email protected]

Matt Taggart writes...

OK, since I have a mostly-working nfsroot and a cross-built toolchain I tried running gcc. I just tried "gcc -o hello hellow.c" where hellow.c is the hello world program from cvs->userspace. The compiler dies with signal 6(abort). The system is also pretty erratic, sometimes the prompt takes a few minutes to return.(once with an out-of-memory error)

Someone suggested that I try an older kernel from before the stack changes were made eariler this month, so I tried an old kernel I had around(built with an old xc). Now gcc gets a little further, but dies with,

------------------------------------------------------------------------------
do_page_fault() pid=18 command='ld'

YZrvWESTHLNXBCVMcbcbcbcbOGFRQPDI
PSW: 00000000000001001111111100001011

r0-3     00000000 fffffff4 000368d7 00107060
r4-7     00000005 000ef790 00125368 00000000
r8-11    000e3438 00000000 00105ca0 00000031
r12-15   000d9ea8 00000000 00107060 000e3120
r16-19   000fa0c0 00106838 00000000 00000000
r20-23   001067e8 00000000 0002cc50 00107060
r24-27   000fa0c0 000d9ea8 000e3120 000d7770
r28-31   00105ca0 00000002 20020500 000368d7
sr0-4    00000000 00000004 00000000 00000004
sr4-8    00000004 00000004 00000004 00000004

IASQ: 00000004 00000004 IAOQ: 0002d5bf 0002d5c3
IIR: 0e752087 ISR: 00000004 IOR: 00000000
ORIG_R28: 00000000
collect2: ld terminated with signal 11 [Segmentation fault], core dumped
------------------------------------------------------------------------------

I also wanted to try the older kernel bits built with the new xc but I couldn't remember the -ffunction-section magic to make it work.

If other people want to play with this I have put my new nfsroots at,

ftp://puffin.external.hp.com/pub/parisc/debug/base-20001227.tar.gz [25MB] ftp://puffin.external.hp.com/pub/parisc/debug/nfsroot-20001227.tar.gz [59MB]

Matt Taggart
[email protected]

What is a newer kernel ?
Date, version ?
What does IIRC mean ?

Ad tool-chain:

I use the tool-chain of the puffingroup, selfmade from the latest version of 21.12.2000 or so. I build it from source and it was no problem except the fact, that the it must be --host' not --HOST' (I have some experiences in it).

But it was not possible to build bash from the scratch in this way. host=hppa-linux, build=i386-linux should be the correct one, but ....
The biuld process of bash (and I think for other too, does not call the cross compiler). The use of CC= AS= LD= etc has the other problem, that binary tools which are build in the build process are running on PA RISC, but they are used in i386 for further building.

So my question: How can I tell an autoconfigured build process (standard linux source tree) to use the cross compiler on the right points.

As I used RedHat up to now, I have no experience in debian packaging system. What are this files like dpkg-cross tool, etc...
How to use them, how to install them.

You say, the bash is a problem. Is this for other tools much easier, like `init' or daemons, etc...

Thanks for the help and a happy new year soon....

Christoph Plattner

-------------------------------------------------------------------------
private:	[email protected]
company:	[email protected]