Re: [OpenBIOS] PReP PowerPC "emulation"

[Massimo Montecchi]

Until r828 was in this way ... then somethink was wrong....

2010/9/13 Laurent Vivier
> Hi,
>
> the last time I tried to boot AIX (6 or 7 monthes ago...), OpenBIOS was
> able to load the bootloader but as RTAS is not provided by OpenBIOS,
> bootloader failed and didn't go further. I think it was with the CHRP
> emulation.
>
> Regards,
> Laurent
>
> Hi,
>
> I have tried qemu from 0.11 to latest git (as of 11/09/2010) with latest
> openbios inside, but if i want to use PREP PPC "machine" the error is the
> same:
>
> qemu: hardware error: PowerPC 601 / 620 / 970 need a 1MB BIOS
>
> CPU #0:
> NIP 00000000 LR 00000000 CTR 00000000 XER 00000000
> MSR 00000000 HID0 00000000 HF 00000000 idx 0
> TB 00000000 00000000 DECR ffffffff
> GPR00 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR04 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR08 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR12 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR16 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR20 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR24 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> GPR28 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> CR 00000000 [ - - - - - - - - ] RES 00000000
> FPR00 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR04 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR08 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR12 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR16 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR20 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR24 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPR28 0000000000000000 0000000000000000 0000000000000000 0000000000000000
> FPSCR 00000000
> SRR0 00000000 SRR1 00000000 SDR1 00000000
>
> How can I use openbios to start an AIX with qemu ?
>
> Massimo Montecchi
> Modena - Italy

PReP PowerPC "emulation"

[Mark Cave-Ayland]

Massimo Montecchi wrote:
> Until r828 was in this way ... then somethink was wrong....

Thanks for the SVN revision information :)

So the main thing that changed in r828 was the handling of the bootargs 
and bootdevice properties to support multiple boot devices. This 
suggests that it could be related to the bootloader either finding a 
value it doesn't like in one of the boot properties, or the new boot 
order doesn't return correctly on invalid input.

If you load qemu with -prom-env 'auto-boot?=false', you should be able 
to manually come up with a "boot <device>" line that will invoke the AIX 
bootloader - any chance you could find out what it is? Also, what format 
is the AIX bootloader - is it an ELF executable?

ATB,

Mark.

PReP PowerPC "emulation"

[Massimo Montecchi]

I know the -prom-env 'auto-boot?=false', but from the openbios "console" I
was not able to run nothing (if you can help me, please!!!)
AFAIK the AIX bootloader is an ELF executable (see here
http://www.ibm.com/developerworks/power/library/pa-spec12/ )

Ciao

2010/9/14 Mark Cave-Ayland
> Massimo Montecchi wrote:
>
> Until r828 was in this way ... then somethink was wrong....
> >
>
> Thanks for the SVN revision information :)
>
> So the main thing that changed in r828 was the handling of the bootargs and
> bootdevice properties to support multiple boot devices. This suggests that
> it could be related to the bootloader either finding a value it doesn't like
> in one of the boot properties, or the new boot order doesn't return
> correctly on invalid input.
>
> If you load qemu with -prom-env 'auto-boot?=false', you should be able to
> manually come up with a "boot <device>" line that will invoke the AIX
> bootloader - any chance you could find out what it is? Also, what format is
> the AIX bootloader - is it an ELF executable?
>
> ATB,
>
> Mark.
>
> Sirius Labs: http://www.siriusit.co.uk/labs

PReP PowerPC "emulation"

[Mark Cave-Ayland]

Massimo Montecchi wrote:
> I know the -prom-env 'auto-boot?=false', but from the openbios "console" 
> I was not able to run nothing (if you can help me, please!!!)
> AFAIK the AIX bootloader is an ELF executable (see 
> here http://www.ibm.com/developerworks/power/library/pa-spec12/ )
>
> Ciao

Is the ELF executable on the CD? You can navigate around the CD using 
the dir command, e.g.

dir cd:,\
dir cd:,\ppc

etc. etc.

Then once you find the ELF executable, try and load it with:

load cd:,\ppc\path\to\file.elf
go

HTH,

Mark.

PReP PowerPC "emulation"

[Andreas Färber]

Am 14.09.2010 um 11:36 schrieb Mark Cave-Ayland:
> If you load qemu with -prom-env 'auto-boot?=false', you should be  
> able to manually come up with a "boot <device>" line that will  
> invoke the AIX bootloader - any chance you could find out what it is?

qemu-system-ppc64 w/ either stock QEMU OpenBIOS or r862 with OSX host  
and ppc MMU patches:

C>> annot manage 'OHCI USB controller' PCI device type 'usb':
> > 106b 3f (c 3 10)
> > =============================================================
> > OpenBIOS 1.0 [Aug 17 2010 14:41]
> > Configuration device id QEMU version 1 machine id 3
> > CPUs: 1
> > Memory: 128M
> > UUID: 00000000-0000-0000-0000-000000000000
> > CPU type PowerPC,970FX

Welcome to OpenBIOS v1.0 built on Aug 17 2010 14:41

0 > load cd:,\ppc\chrp\bootfile.exe  ok
0 > go No valid state has been set by load or init-program
  ok
0 > boot cd:,\ppc\chrp\bootfile.exe No valid state has been set by  
load or init-program
  ok
0 > dir cd:,\ppc\chrp
       2048 2007-07-03 17:39:44 .\
       2048 2007-07-03 17:32:05 ..\
   12487680 2007-07-03 17:39:24 bootfile.exe
  ok
0 >
> Also, what format is the AIX bootloader - is it an ELF executable?

$ file /Volumes/CDROM/ppc/chrp/bootfile.exe
/Volumes/CDROM/ppc/chrp/bootfile.exe: ELF 32-bit MSB executable,  
PowerPC or cisco 4500, version 1 (SYSV), statically linked, corrupted  
section header size

Andreas

PReP PowerPC "emulation"

[Mark Cave-Ayland]

Andreas Färber wrote:
> qemu-system-ppc64 w/ either stock QEMU OpenBIOS or r862 with OSX host 
> and ppc MMU patches:
>
> C>> annot manage 'OHCI USB controller' PCI device type 'usb':
> > > 106b 3f (c 3 10)
> > > =============================================================
> > > OpenBIOS 1.0 [Aug 17 2010 14:41]
> > > Configuration device id QEMU version 1 machine id 3
> > > CPUs: 1
> > > Memory: 128M
> > > UUID: 00000000-0000-0000-0000-000000000000
> > > CPU type PowerPC,970FX
>
> Welcome to OpenBIOS v1.0 built on Aug 17 2010 14:41
>
> 0 > load cd:,\ppc\chrp\bootfile.exe  ok
> 0 > go No valid state has been set by load or init-program
>  ok
> 0 > boot cd:,\ppc\chrp\bootfile.exe No valid state has been set by load 
> or init-program
>  ok
> 0 > dir cd:,\ppc\chrp
>       2048 2007-07-03 17:39:44 .\
>       2048 2007-07-03 17:32:05 ..\
>   12487680 2007-07-03 17:39:24 bootfile.exe
>  ok
> 0 >
> > Also, what format is the AIX bootloader - is it an ELF executable?
>
> $ file /Volumes/CDROM/ppc/chrp/bootfile.exe
> /Volumes/CDROM/ppc/chrp/bootfile.exe: ELF 32-bit MSB executable, PowerPC 
> or cisco 4500, version 1 (SYSV), statically linked, corrupted section 
> header size
>
> Andreas

Right. So in the case that you see the message "No valid state has been 
set by load or init-program", this means that either the load failed or 
init-program failed to detect a valid ELF header for the current 
architecture.

First thing to check is that the executable is being loaded from disk, e.g.

load cd:,\ppc\chrp\bootfile.exe

then dump the first 200 bytes of memory to make sure an ELF header is 
present (i.e. the load from disk was successful):

load-base 200 dump

If the ELF header is present then it must be init-program which is 
failing. If the ELF header is not present, you'll need to take a look in 
either libopenbios/load.c and/or libopenbios/elf_load.c at elf_load() 
and elf_init_program().

The one thing that I did notice is that you are attempting to launch a 
PPC32 ELF file under a PPC64 Qemu, so perhaps it is the checks in 
is_elf() which are failing? Try taking a look at the relevant constants 
in include/arch/ppc/elf.h.

HTH,

Mark.

PReP PowerPC "emulation"

[Massimo Montecchi]

Mark, using your hints I was able to establish that the ELF header is not
present.
I will try (I'm not so expert in C) to take a look at the files you suggest
me.
Thanks, Ciao

2010/9/15 Mark Cave-Ayland
> Andreas Färber wrote:
>
> qemu-system-ppc64 w/ either stock QEMU OpenBIOS or r862 with OSX host and
> > ppc MMU patches:
> >
> > C>> annot manage 'OHCI USB controller' PCI device type 'usb':
> > > > 106b 3f (c 3 10)
> > > > =============================================================
> > > > OpenBIOS 1.0 [Aug 17 2010 14:41]
> > > > Configuration device id QEMU version 1 machine id 3
> > > > CPUs: 1
> > > > Memory: 128M
> > > > UUID: 00000000-0000-0000-0000-000000000000
> > > > CPU type PowerPC,970FX
> >
> > Welcome to OpenBIOS v1.0 built on Aug 17 2010 14:41
> >
> > 0 > load cd:,\ppc\chrp\bootfile.exe  ok
> > 0 > go No valid state has been set by load or init-program
> >  ok
> > 0 > boot cd:,\ppc\chrp\bootfile.exe No valid state has been set by load or
> > init-program
> >  ok
> > 0 > dir cd:,\ppc\chrp
> >      2048 2007-07-03 17:39:44 .\
> >      2048 2007-07-03 17:32:05 ..\
> >  12487680 2007-07-03 17:39:24 bootfile.exe
> >  ok
> > 0 >
> >
> > Also, what format is the AIX bootloader - is it an ELF executable?
> > >
> >
> > $ file /Volumes/CDROM/ppc/chrp/bootfile.exe
> > /Volumes/CDROM/ppc/chrp/bootfile.exe: ELF 32-bit MSB executable, PowerPC
> > or cisco 4500, version 1 (SYSV), statically linked, corrupted section header
> > size
> >
> > Andreas
>
> Right. So in the case that you see the message "No valid state has been set
> by load or init-program", this means that either the load failed or
> init-program failed to detect a valid ELF header for the current
> architecture.
>
> First thing to check is that the executable is being loaded from disk, e.g.
>
> load cd:,\ppc\chrp\bootfile.exe
>
> then dump the first 200 bytes of memory to make sure an ELF header is
> present (i.e. the load from disk was successful):
>
> load-base 200 dump
>
> If the ELF header is present then it must be init-program which is failing.
> If the ELF header is not present, you'll need to take a look in either
> libopenbios/load.c and/or libopenbios/elf_load.c at elf_load() and
> elf_init_program().
>
> The one thing that I did notice is that you are attempting to launch a
> PPC32 ELF file under a PPC64 Qemu, so perhaps it is the checks in is_elf()
> which are failing? Try taking a look at the relevant constants in
> include/arch/ppc/elf.h.
>
> HTH,
>
> Mark.
>
> Sirius Labs: http://www.siriusit.co.uk/labs

PReP PowerPC "emulation"

[Andreas Färber]

Am 15.09.2010 um 11:06 schrieb Mark Cave-Ayland:
> Andreas Färber wrote:
> > qemu-system-ppc64 w/ either stock QEMU OpenBIOS or r862 with OSX  
> > host and ppc MMU patches:
> > C>> annot manage 'OHCI USB controller' PCI device type 'usb':
> > > > 106b 3f (c 3 10)
> > > >
> > > > OpenBIOS 1.0 [Aug 17 2010 14:41]
> > > > Configuration device id QEMU version 1 machine id 3
> > > > CPUs: 1
> > > > Memory: 128M
> > > > UUID: 00000000-0000-0000-0000-000000000000
> > > > CPU type PowerPC,970FX
> >
> > Welcome to OpenBIOS v1.0 built on Aug 17 2010 14:41
> > 0 > load cd:,\ppc\chrp\bootfile.exe  ok
> > 0 > go No valid state has been set by load or init-program
> > ok
> > 0 > boot cd:,\ppc\chrp\bootfile.exe No valid state has been set by  
> > load or init-program
> > ok
> > 0 > dir cd:,\ppc\chrp
> >      2048 2007-07-03 17:39:44 .\
> >      2048 2007-07-03 17:32:05 ..\
> >  12487680 2007-07-03 17:39:24 bootfile.exe
> > ok
> > 0 >
> > > Also, what format is the AIX bootloader - is it an ELF executable?
> >
> > $ file /Volumes/CDROM/ppc/chrp/bootfile.exe
> > /Volumes/CDROM/ppc/chrp/bootfile.exe: ELF 32-bit MSB executable,  
> > PowerPC or cisco 4500, version 1 (SYSV), statically linked,  
> > corrupted section header size
> > Andreas
>
> Right. So in the case that you see the message "No valid state has  
> been set by load or init-program", this means that either the load  
> failed or init-program failed to detect a valid ELF header for the  
> current architecture.
>
> First thing to check is that the executable is being loaded from  
> disk, e.g.
>
> load cd:,\ppc\chrp\bootfile.exe
>
> then dump the first 200 bytes of memory to make sure an ELF header  
> is present (i.e. the load from disk was successful):
>
> load-base 200 dump
>
> If the ELF header is present then it must be init-program which is  
> failing. If the ELF header is not present, you'll need to take a  
> look in either libopenbios/load.c and/or libopenbios/elf_load.c at  
> elf_load() and elf_init_program().
>
> The one thing that I did notice is that you are attempting to launch  
> a PPC32 ELF file under a PPC64 Qemu, so perhaps it is the checks in  
> is_elf() which are failing? Try taking a look at the relevant  
> constants in include/arch/ppc/elf.h.

If I can trust printk, I do not arrive in the C load() function  
(libopenbios/load.c) at all.
According to `debug open-dev`, path-resolution returns 0 for "load cd:, 
\ppc\chrp\bootfile.exe" (forth/device/pathres.fs).

Any ideas?

Andreas

PReP PowerPC "emulation"

[Andreas Färber]

The following patch does wonders, it changes the ppc config to match  
sparc64 wrt ISO9660:

diff --git a/config/examples/cross-ppc_config.xml b/config/examples/ 
cross-ppc_config.xml
index d658ac4..c180509 100644
--- a/config/examples/cross-ppc_config.xml
+++ b/config/examples/cross-ppc_config.xml
@@ -55,7 +55,7 @@
    <option name="CONFIG_PC_PARTS" type="boolean" value="true"/>
    <option name="CONFIG_HFS" type="boolean" value="true"/>
    <option name="CONFIG_HFSP" type="boolean" value="true"/>
-  <option name="CONFIG_ISO9660" type="boolean" value="true"/>
+  <option name="CONFIG_ISO9660" type="boolean" value="false"/>
    <option name="CONFIG_EXT2" type="boolean" value="true"/>
    <option name="CONFIG_GRUBFS" type="boolean" value="true"/>
    <option name="CONFIG_FSYS_EXT2FS" type="boolean" value="false"/>
@@ -65,7 +65,7 @@
    <option name="CONFIG_FSYS_REISERFS" type="boolean" value="false"/>
    <option name="CONFIG_FSYS_XFS" type="boolean" value="false"/>
    <option name="CONFIG_FSYS_UFS" type="boolean" value="false"/>
-  <option name="CONFIG_FSYS_ISO9660" type="boolean" value="false"/>
+  <option name="CONFIG_FSYS_ISO9660" type="boolean" value="true"/>
    <option name="CONFIG_FSYS_FFS" type="boolean" value="false"/>
    <option name="CONFIG_FSYS_VSTAFS" type="boolean" value="false"/>
    <option name="CONFIG_FSYS_NTFS" type="boolean" value="false"/>

$ qemu-system-ppc64 -boot d -cdrom path/to/6.1.0-dvd.iso -nographic

C>> annot manage 'OHCI USB controller' PCI device type 'usb':
> > 106b 3f (c 3 10)
> > =============================================================
> > OpenBIOS 1.0 [Sep 25 2010 23:55]
> > Configuration device id QEMU version 1 machine id 3
> > CPUs: 1
> > Memory: 128M
> > UUID: 00000000-0000-0000-0000-000000000000
> > CPU type PowerPC,970FX

Welcome to OpenBIOS v1.0 built on Sep 25 2010 23:55
Trying cd:,\:tbxi...
File not found
Trying cd:,\ppc\bootinfo.txt...

-------------------------------------------------------------------------------
                                 Welcome to AIX.
                        boot image timestamp: 01:75 2B/75
NULL ihandle                 The current time and date: 00:00:00  
156828/00/0008
          processor count: 0;  memory size: 128MB;  kernel size: 8391752
                     boot device: cd:\ppc\chrp\bootfile.exe
NULL phandle
Unexpected client interface exception: -1
NULL phandle
Unexpected client interface exception: -1

ERROR: This system is incompatible with the 64 bit kernel.

EXIT
0 >

`dir cd:,\ppc\chrp` is unsupported in that configuration though.

Andreas

PReP PowerPC "emulation"

[Mark Cave-Ayland]

Andreas Färber wrote:
> The following patch does wonders, it changes the ppc config to match 
> sparc64 wrt ISO9660:

Ah wait a second - I think I know what's happening here. Several months 
ago I found a couple of bugs in fs/iso9660/* related to loading files 
from CD. I seem to recall:

i) The filenames all had to be in capital letters
ii) Sometimes I had to put an extra . at the end

I found these bugs by stepping through iso9660_opendir() under gdb. So 
that is likely what's happening here - OpenBIOS is unable to locate the 
file from the filesystem at boot time.
> `dir cd:,\ppc\chrp` is unsupported in that configuration though.

Yeah because while the grubfs iso9660 implementation doesn't have the 
case-sensitivity/period bugs, unfortunately the grub API doesn't provide 
a function to list files within a directory.

If you could provide a patch to fix these bugs in fs/iso9660/* then I 
would definitely apply it :)

ATB,

Mark.

PReP PowerPC "emulation"

[Andreas Färber]

Am 26.09.2010 um 14:12 schrieb Mark Cave-Ayland:
> Andreas Färber wrote:
> > The following patch does wonders, it changes the ppc config to  
> > match sparc64 wrt ISO9660:
>
> Ah wait a second - I think I know what's happening here. Several  
> months ago I found a couple of bugs in fs/iso9660/* related to  
> loading files from CD. I seem to recall:
>
> i) The filenames all had to be in capital letters
> ii) Sometimes I had to put an extra . at the end
>
> I found these bugs by stepping through iso9660_opendir() under gdb.  
> So that is likely what's happening here - OpenBIOS is unable to  
> locate the file from the filesystem at boot time.

iso9660_open() actually succeeds for lowercase \ppc\chrp\bootfile.exe.  
Replacing strcmp() with strcasecmp() in seek_name() made no visible  
difference. It probably depends on the case on the disk, some  
directories are uppercase here, some lowercase.

diff --git a/fs/iso9660/iso9660_opendir.c b/fs/iso9660/iso9660_opendir.c
index 10cb89d..49eab4b 100644
--- a/fs/iso9660/iso9660_opendir.c
+++ b/fs/iso9660/iso9660_opendir.c
@@ -55,7 +55,7 @@ static struct iso_directory_record *  
seek_name(iso9660_VOLUME *volume,
         while ((idr = iso9660_readdir(dir)) != NULL)
         {
                 iso9660_name(volume, idr, name_buf);
-               if (strcmp(name, name_buf) == 0)
+               if (strcasecmp(name, name_buf) == 0)
                 {
                         result = idr_new(idr);
                         iso9660_closedir(dir);

Some random observations:

* Changing some file in fs/iso9660/ and re-running `make` in obj-ppc  
causes LD to fail with multiple/undefined symbols. Removing ./libfs.a  
first and then re-running `make` works fine.

* The ofmem_claim() hack I tried for Haiku does not help with AIX. AIX  
tries to claim 0x00200000 bytes at location 0x07e00000, which overlaps  
with OpenBIOS at rom_base()==0x07f00000. Haiku by comparison claims  
0x00100000 bytes at 0x0 and ended up at a different location by my  
patch (0x07e00000 vs. 0x07f00000).

* The two AIX "NULL phandle" messages go away and the processor count  
is detected as 1 when running with -m 1024:

-------------------------------------------------------------------------------
                                 Welcome to AIX.
                        boot image timestamp: 01:75 2B/75
NULL ihandle                 The current time and date: 00:00:00  
156828/00/0008
         processor count: 1;  memory size: 1024MB;  kernel size: 8391752
                     boot device: cd:\ppc\chrp\bootfile.exe
Validation failed: the "/rtas" device node does not exist.
EXIT
0 >

The "NULL ihandle" stems from call-method being called with 0x0  
ihandle for get-time. This is caused by open for the rtc node  
returning 0x0 on ppc64. ppc output follows:

-------------------------------------------------------------------------------
                                 Welcome to AIX.
                        boot image timestamp: 01:75 2B/75
                  The current time and date: 18:08:20 09/26/2010
         processor count: 1;  memory size: 1024MB;  kernel size: 8391752
                     boot device: cd:\ppc\chrp\bootfile.exe

ERROR: This system is not supported for use with AIX 6.1.
              model: Power Macintosh        processor: PowerPC,750

AIX 6.1 requires the POWER4 (or later) processor.

EXIT
0 >

Andreas

libfs.a issue (was: PReP PowerPC "emulation")

[Andreas Färber]

Am 26.09.2010 um 20:53 schrieb Andreas Färber:
> Changing some file in fs/iso9660/ and re-running `make` in obj-ppc
>
> causes LD to fail with multiple/undefined symbols. Removing ./ 
> libfs.a first and then re-running `make` works fine.

I believe this is caused by identically named .o files in target/fs/ 
hfs/ and target/fs/hfsplus/. It therefore doesn't affect the sparc  
targets.
Should we rename the sources or always rm the archive first?

Andreas

libfs.a issue (was: PReP PowerPC "emulation")

[Blue Swirl]

On Mon, Oct 11, 2010 at 2:32 PM, Andreas Färber  wrote:
> Am 26.09.2010 um 20:53 schrieb Andreas Färber:
> > Changing some file in fs/iso9660/ and re-running `make` in obj-ppc
> >
> > causes LD to fail with multiple/undefined symbols. Removing ./libfs.a first
> > and then re-running `make` works fine.
>
> I believe this is caused by identically named .o files in target/fs/hfs/ and
> target/fs/hfsplus/. It therefore doesn't affect the sparc targets.
> Should we rename the sources or always rm the archive first?

Rename, but does it actually make sense to enable both HFS and HFS+
support simultaneously? Is HFS+ able to read HFS volumes?