[PATCH] First support for HP DL145 G3

[Mondrian Nuessle]

Hi everybody,

I prepared an initial set of patches to support the HP DL145 G3 server, which is a Broadcom HT-1000/HT-2100 based Socket
F machine.

The current status of the port is:

* we have serial output (using the serverengines pilot chip on the board, see attached pilot_superio.patch)
* CPU and memory gets initialized
* on-board eth0 is working
* SATA is working
* the on-board graphic was not tested (not important for
  a server and the graphics sucks anyway...)
* USB and PS2 work
* no ACPI
* I remember that I could also use a PCIe card, see attached bcm21000.patch
* HTX card does work.
* Earlier problems we had in respect to timer interrupt are not there anymore.
* the hp_dl145_g3.patch includes the mainboard and the target dirs

BUT:
* if I coldboot, the CPUs are only running at 1 GHz and not with 2.6 GHz as they should...
* if we boot the factory BIOS and then perform a warm-reset into Coreboot, the CPUs run with 2.6 GHz.

I hope someone on the list can help.

* There are also some more problems with the HT-1000 (aka bcm5785) southbridge:
  - if I activate IDE in bcm5785_early_setup.c, the machine will freeze shortly after being completely booted
  - if I activate the second IOAPIC (line 179 in bcm5785_early_setup.c), my linux boot hangs when starting udevd...
  - with both features disabled (see attached bcm5785_noide.patch), the machine boots and does not crash, of course
    PATA drives are not available.

To boot the machine, I tested FILO. The on-board SATA adapter uses a class id of 0x104 in sata mode
(which we use). FILO does not accept this class ID; I prepared a patch that adds this class ID to
the respective if statement in driver/ide.c. FILO does then boot from SATA without any problems (see attached
filo_dl145_sata.patch).

I hope this gets you started Samuel, and I hope the list can provide some insight regarding the CPU
frequency problem.

Regards,
Mondrian

[samuel]

I just tried building this :)

But I didn't succeed.
What I did:
1. checkout payload filo, apply filo_dl_145_sata.patch and build it.
svn co svn://coreboot.org/filo
cd filo/trunk/filo
patch -p 0 << ~/coreboot/filo_dl_145_sata.patch
sh build.sh
cp build/filo.elf ~/coreboot/

2. checkout coreboot-v2, apply remaining patches and build it:
svn co svn://coreboot.org/repos/trunk/coreboot-v2
cd coreboot-v2
patch -p 0 << ~/coreboot/bcm2100.patch
patch -p 0 << ~/coreboot/pilot_superio.patch
patch -p 0 << ~/coreboot/hp_dl145_g3.patch
patch -p 0 << ~/bcm5785_noide.patch
cd targets
buildtarget hp/dl145_g3 /home/samuel/coreboot/coreboot_v2
cd hp/dl145_g3
sed -e s@/home/mondrian/filo.elf@/home/samuel/coreboot/filo.elf@g Config.lb
cd dl145_g3
make

and then i get the error:
http://merlin.ugent.be/~samuel/dl145g3/builds/firmware_build_20090331_1708.t...

Any suggesions? Did I do something wrong?

2009/3/31 Mondrian Nuessle :
> Hi everybody,
>
> I prepared an initial set of patches to support the HP DL145 G3 server, which is a Broadcom HT-1000/HT-2100 based Socket
> F machine.
>
> The current status of the port is:
> we have serial output (using the serverengines pilot chip on the board, see attached pilot_superio.patch)
> CPU and memory gets initialized
> on-board eth0 is working
> SATA is working
> the on-board graphic was not tested (not important for
>
> a server and the graphics sucks anyway...)
> USB and PS2 work
> no ACPI
> I remember that I could also use a PCIe card, see attached bcm21000.patch
> HTX card does work.
> Earlier problems we had in respect to timer interrupt are not there anymore.
> the hp_dl145_g3.patch includes the mainboard and the target dirs
>
> BUT:
> if I coldboot, the CPUs are only running at 1 GHz and not with 2.6 GHz as they should...
> if we boot the factory BIOS and then perform a warm-reset into Coreboot, the CPUs run with 2.6 GHz.
>
> I hope someone on the list can help.
> There are also some more problems with the HT-1000 (aka bcm5785) southbridge:
>
> - if I activate IDE in bcm5785_early_setup.c, the machine will freeze shortly after being completely booted
>  - if I activate the second IOAPIC (line 179 in bcm5785_early_setup.c), my linux boot hangs when starting udevd...
>  - with both features disabled (see attached bcm5785_noide.patch), the machine boots and does not crash, of course
>    PATA drives are not available.
>
> To boot the machine, I tested FILO. The on-board SATA adapter uses a class id of 0x104 in sata mode
> (which we use). FILO does not accept this class ID; I prepared a patch that adds this class ID to
> the respective if statement in driver/ide.c. FILO does then boot from SATA without any problems (see attached
> filo_dl145_sata.patch).
>
> I hope this gets you started Samuel, and I hope the list can provide some insight regarding the CPU
> frequency problem.
>
> Regards,
> Mondrian

[Myles Watson]

On Tue, Mar 31, 2009 at 10:40 AM, samuel  wrote:
> I just tried building this :)
>
> But I didn't succeed.
> What I did:
> checkout payload filo, apply filo_dl_145_sata.patch and build it.
>
> svn co svn://coreboot.org/filo
> cd filo/trunk/filo
> patch -p 0 << ~/coreboot/filo_dl_145_sata.patch
> sh build.sh
> cp build/filo.elf ~/coreboot/
> checkout coreboot-v2, apply remaining patches and build it:
>
> svn co svn://coreboot.org/repos/trunk/coreboot-v2
> cd coreboot-v2
> patch -p 0 << ~/coreboot/bcm2100.patch
> patch -p 0 << ~/coreboot/pilot_superio.patch
> patch -p 0 << ~/coreboot/hp_dl145_g3.patch
> patch -p 0 << ~/bcm5785_noide.patch
> cd targets
> buildtarget hp/dl145_g3 /home/samuel/coreboot/coreboot_v2
> cd hp/dl145_g3
> sed -e s@/home/mondrian/filo.elf@/home/samuel/coreboot/filo.elf@g Config.lb
> cd dl145_g3
> make
>
> and then i get the error:

cp coreboot_ram.nrv2b coreboot_ram.rom
echo '/*ldoptions*/' > ldscript.ld; cat ldoptions >> ldscript.ld ; for
file in  /home/samuel/coreboot/coreboot-v2/src/arch/i386/init/ldscript_fallback.lb
/home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/entry16.lds
/home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/reset16.lds
/home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/id.lds
/home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/failover.lds ; do
echo /* $file */ >> ldscript.ld; cat $file >> ldscript.ld ; done
gcc -m32 -Wl,--build-id=none -nostdlib -nostartfiles -static -o
coreboot -T ldscript.ld crt0.o
/usr/bin/ld: coreboot: section `.id' can't be allocated in segment 2
/usr/bin/ld: final link failed: Bad value
collect2: ld returned 1 exit status
> Any suggesions? Did I do something wrong?

Try a different version of binutils and/or gcc.  I used to get this
error on Fedora Core 8.

Thanks,
Myles

[samuel]

On Tue, Mar 31, 2009 at 6:56 PM, Myles Watson  wrote:
> On Tue, Mar 31, 2009 at 10:40 AM, samuel  wrote:
> > I just tried building this :)
> >
> > But I didn't succeed.
> > What I did:
> > checkout payload filo, apply filo_dl_145_sata.patch and build it.
> >
> > svn co svn://coreboot.org/filo
> > cd filo/trunk/filo
> > patch -p 0 << ~/coreboot/filo_dl_145_sata.patch
> > sh build.sh
> > cp build/filo.elf ~/coreboot/
> > checkout coreboot-v2, apply remaining patches and build it:
> >
> > svn co svn://coreboot.org/repos/trunk/coreboot-v2
> > cd coreboot-v2
> > patch -p 0 << ~/coreboot/bcm2100.patch
> > patch -p 0 << ~/coreboot/pilot_superio.patch
> > patch -p 0 << ~/coreboot/hp_dl145_g3.patch
> > patch -p 0 << ~/bcm5785_noide.patch
> > cd targets
> > buildtarget hp/dl145_g3 /home/samuel/coreboot/coreboot_v2
> > cd hp/dl145_g3
> > sed -e s@/home/mondrian/filo.elf@/home/samuel/coreboot/filo.elf@g Config.lb
> > cd dl145_g3
> > make
> >
> > and then i get the error:
>
> cp coreboot_ram.nrv2b coreboot_ram.rom
> echo '/*ldoptions*/' > ldscript.ld; cat ldoptions >> ldscript.ld ; for
> file in  /home/samuel/coreboot/coreboot-v2/src/arch/i386/init/ldscript_fallback.lb
> /home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/entry16.lds
> /home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/reset16.lds
> /home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/id.lds
> /home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/failover.lds ; do
> echo /* $file */ >> ldscript.ld; cat $file >> ldscript.ld ; done
> gcc -m32 -Wl,--build-id=none -nostdlib -nostartfiles -static -o
> coreboot -T ldscript.ld crt0.o
> /usr/bin/ld: coreboot: section `.id' can't be allocated in segment 2
> /usr/bin/ld: final link failed: Bad value
> collect2: ld returned 1 exit status
> > Any suggesions? Did I do something wrong?
>
> Try a different version of binutils and/or gcc.  I used to get this
> error on Fedora Core 8.

samuel@camelot ~ $ lvu version binutils
2.18
samuel@camelot ~ $ lvu version gcc
4.2.4

You suggest downgrading or upgrading? :)
> Thanks,
> Myles

[Myles Watson]

On Tue, Mar 31, 2009 at 10:59 AM, samuel  wrote:
> On Tue, Mar 31, 2009 at 6:56 PM, Myles Watson  wrote:
> > On Tue, Mar 31, 2009 at 10:40 AM, samuel  wrote:
> > > I just tried building this :)
> > >
> > > But I didn't succeed.
> > > What I did:
> > > checkout payload filo, apply filo_dl_145_sata.patch and build it.
> > >
> > > svn co svn://coreboot.org/filo
> > > cd filo/trunk/filo
> > > patch -p 0 << ~/coreboot/filo_dl_145_sata.patch
> > > sh build.sh
> > > cp build/filo.elf ~/coreboot/
> > > checkout coreboot-v2, apply remaining patches and build it:
> > >
> > > svn co svn://coreboot.org/repos/trunk/coreboot-v2
> > > cd coreboot-v2
> > > patch -p 0 << ~/coreboot/bcm2100.patch
> > > patch -p 0 << ~/coreboot/pilot_superio.patch
> > > patch -p 0 << ~/coreboot/hp_dl145_g3.patch
> > > patch -p 0 << ~/bcm5785_noide.patch
> > > cd targets
> > > buildtarget hp/dl145_g3 /home/samuel/coreboot/coreboot_v2
> > > cd hp/dl145_g3
> > > sed -e s@/home/mondrian/filo.elf@/home/samuel/coreboot/filo.elf@g Config.lb
> > > cd dl145_g3
> > > make
> > >
> > > and then i get the error:
> >
> > cp coreboot_ram.nrv2b coreboot_ram.rom
> > echo '/*ldoptions*/' > ldscript.ld; cat ldoptions >> ldscript.ld ; for
> > file in  /home/samuel/coreboot/coreboot-v2/src/arch/i386/init/ldscript_fallback.lb
> > /home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/entry16.lds
> > /home/samuel/coreboot/coreboot-v2/src//cpu/x86/16bit/reset16.lds
> > /home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/id.lds
> > /home/samuel/coreboot/coreboot-v2/src//arch/i386/lib/failover.lds ; do
> > echo /* $file */ >> ldscript.ld; cat $file >> ldscript.ld ; done
> > gcc -m32 -Wl,--build-id=none -nostdlib -nostartfiles -static -o
> > coreboot -T ldscript.ld crt0.o
> > /usr/bin/ld: coreboot: section `.id' can't be allocated in segment 2
> > /usr/bin/ld: final link failed: Bad value
> > collect2: ld returned 1 exit status
> > > Any suggesions? Did I do something wrong?
> >
> > Try a different version of binutils and/or gcc.  I used to get this
> > error on Fedora Core 8.
>
> samuel@camelot ~ $ lvu version binutils
> 2.18
> samuel@camelot ~ $ lvu version gcc
> 4.2.4
>
> You suggest downgrading or upgrading? :)

I wish I knew.  With Fedora I couldn't do either.  I just had to
switch distros.  Interestingly, it only affected my 32-bit machine;
64-bit was fine.

Now I use:

gcc (Ubuntu 4.3.2-1ubuntu12) 4.3.2
GNU assembler (GNU Binutils for Ubuntu) 2.18.93.20081009

I know there are others that work too.

Thanks,
Myles

[Stefan Reinauer]

On 31.03.2009 16:01 Uhr, Mondrian Nuessle wrote:
> I prepared an initial set of patches to support the HP DL145 G3 server, which is a Broadcom HT-1000/HT-2100 based Socket
> F machine.
> I remember that I could also use a PCIe card, see attached bcm21000.patch

Hi Mondrian,

thank you very much for your patch. Do you happen to know if the component's name is HT2100 or HT21000? Just to make sure we have it in the repository correctly.

Stefan

[Tom Sylla]

On Tue, Mar 31, 2009 at 12:56 PM, Stefan Reinauer  wrote:
> thank you very much for your patch. Do you happen to know if the component's name is HT2100 or HT21000? Just to make sure we have it in the repository correctly.

The naming in the patch looks right. Broadcom refers to it as HT-2100
on their website:

http://www.broadcom.com/products/Small-Medium-Business/SystemI-O-Products/HT...

and as both HT-2100 and BCM21000 in the datasheet (often as "BCM21000
(HT-2100)" ).

[Mondrian Nuessle]

> > thank you very much for your patch. Do you happen to know if the component's name is HT2100 or HT21000? Just to make sure we have it in the repository correctly.
>
> The naming in the patch looks right. Broadcom refers to it as HT-2100
> and as both HT-2100 and BCM21000 in the datasheet (often as "BCM21000
> (HT-2100)" ).

that's my understanding, too.

[samuel]

OK,

I managed booting the firmware... I never worked with serial ports and
stuff so this took me a while to grab the idea...

Anyway,
you can find the boot.log on http://merlin.ugent.be/~samuel/dl145g3/boot.log

Obviously that boot line is wrong.
http://merlin.ugent.be/~samuel/dl145g3/filo_config
obviously CONFIG_MENULST_FILE="hda3:/boot/filo/menu.lst" is wrong. My
kernel I'd like to boot is hda1:/2.6.28.8_x86-64

Trying to boot this as follows:

kernel hda1:/2.6.28.8_x86_64

results in:

Drive 0 does not exist
Error 15: File not found.

I'm a bit stuck I'm afraid, afaik i did patch filo with the patch from
mondrian so prolly me config of filo is wrong?

Kind regards,

Samuel

[samuel]

ok... Fixed that too :)

used the following bootline: kernel hda1:/2.6.28.8-x86_64
root=/dev/md1 console=tty0 console=ttyS0,115200

and that worked like a charm...
Logged in with ssh and it works :D
Even kvm was just working:
from dmesg: loaded kvm module (kvm-84)

just wonderfull

So now on to the "issue list"
1. Cpu's are indeed running at 1Ghz. Running cpufreq-info says: no or
unknown cpufreq driver is active on this CPU

2. not all that important but some vga output would be convenient..
.right now it's doing nothing...

Anyway i need to thank everyone already for helping me gettng this far
:) I hope to get the procs running at full speed and then we can start
virtualzing machine :D

Kind regards,

Samuel

[Marc Jones]

> So now on to the "issue list"
> Cpu's are indeed running at 1Ghz. Running cpufreq-info says: no or
>
> unknown cpufreq driver is active on this CPU

#define K8_SET_FIDVID 0 needs to be a 1 for full speed but there might
need to be some tweaking in cache_as_ram.c where it is used. Compare
it with a known working mainboard.

Marc

[samuel]

On Thu, Apr 2, 2009 at 8:54 PM, Marc Jones  wrote:
> > So now on to the "issue list"
> > Cpu's are indeed running at 1Ghz. Running cpufreq-info says: no or
> >
> > unknown cpufreq driver is active on this CPU
>
> #define K8_SET_FIDVID 0 needs to be a 1 for full speed but there might
> need to be some tweaking in cache_as_ram.c where it is used. Compare
> it with a known working mainboard.

That indeed did the trick. It's running at 2.6Ghz now :)
> Marc

[samuel]

Anyone has any hints on how to get the vga port going?

and another thing...
I'm trying to get filo/grub to autoload my config file i made... i
stored it on /dev/sda1 (together with the kernel) but it is not
loading that file... it simply goes straight to the prompt. No error
nothing...

the config i made looks like this:
#
timeout 10
default 0
fallback 1

title 2.6.28.8-x86_64
kernel /dev/sda1/2.6.28.8-x86_64 root=/dev/md1 console=tty0 console=ttyS0,115200
#

any ideas?

[Myles Watson]

On Thu, Apr 2, 2009 at 2:45 PM, samuel  wrote:
> Anyone has any hints on how to get the vga port going?

I don't see it getting initialized.  I don't see a line where the ROM
is being run in your log.

Have you tried extracting the VGA BIOS?

Do you have a vga card to try?
> and another thing...
> I'm trying to get filo/grub to autoload my config file i made... i
> stored it on /dev/sda1 (together with the kernel) but it is not
> loading that file... it simply goes straight to the prompt. No error
> nothing...
>
> the config i made looks like this:
> #
> timeout 10
> default 0
> fallback 1
>
> title 2.6.28.8-x86_64
> kernel /dev/sda1/2.6.28.8-x86_64 root=/dev/md1 console=tty0 console=ttyS0,115200

/dev/sda1 looks wrong here.  Have you tried hda1 (like your working boot line)?

Myles

[samuel]

On Thu, Apr 2, 2009 at 11:00 PM, Myles Watson  wrote:
> On Thu, Apr 2, 2009 at 2:45 PM, samuel  wrote:
> > Anyone has any hints on how to get the vga port going?
>
> I don't see it getting initialized.  I don't see a line where the ROM
> is being run in your log.
>
> Have you tried extracting the VGA BIOS?
>
> Do you have a vga card to try?

I will try that
> > and another thing...
> > I'm trying to get filo/grub to autoload my config file i made... i
> > stored it on /dev/sda1 (together with the kernel) but it is not
> > loading that file... it simply goes straight to the prompt. No error
> > nothing...
> >
> > the config i made looks like this:
> > #
> > timeout 10
> > default 0
> > fallback 1
> >
> > title 2.6.28.8-x86_64
> > kernel /dev/sda1/2.6.28.8-x86_64 root=/dev/md1 console=tty0 console=ttyS0,115200
>
> /dev/sda1 looks wrong here.  Have you tried hda1 (like your working boot line)?

But shouldn't it at least show the menu? Even if the entry is completely wrong?
> Myles

[samuel]

Hi,

Status of dl145g3:
* 2 Opteron 2218HE's working in the machine. No problem
* both cpu's got 4GB of ram, NUMA is working :)
* booting with filo works
* converted the patches of Mondrian to the last revision of coreboot.
(original patches are on trac, issue 127)
  * src/mainboard/hp/dl145g3/Config.lb:
http://merlin.ugent.be/~samuel/dl145g3/config/srcConfig.lb
  * src/mainboard/hp/dl145g3/Options.lb:
http://merlin.ugent.be/~samuel/dl145g3/config/srcOptions.lb
  * targets/hp/dl145g3/Config.lb:
http://merlin.ugent.be/~samuel/dl145g3/config/Config.lb
* Contacted Anton Borisov to extract the VGA rom of the orginal HP
bios: http://merlin.ugent.be/~samuel/dl145g3/OPROM.00.rom

2 concerns:
1. CPU0 gets very hot during idling. Not really sure what is going on
there. Does the code take into account that this is actually a HE (low
power opteron)? or is that ignored? How can i check that the voltage
applied by coreboot is correct? Maybe it's ok.. I just want a
confirmation on this.
2. VGA output is still not working. I have tried both YABEL and
X86BIOS to load the vga rom but i get errors...
X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
relevant part:
On mainboard, rom address for PCI: 00:04.0 = fff00000
copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
entering emulator
int1a vector at 274af

YABEL: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootYABEL.log
relevant part:
On mainboard, rom address for PCI: 00:04.0 = fff00000
copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
pci_cfg_read(): Config read access invalid device! bus: 00 (00),
devfn: 00 (20), offs: 00

I've also tried with a seperate VGA card on the pci-e slot on the
machine but that doesn't seem to work either.

As no one on IRC seems to be able to help out with this I'd like to
hear from you :)

Cheers,

Samuel

[Myles Watson]

> -----Original Message-----
> From: [e-mail] [mailto:[e-mail]]
> On Behalf Of samuel
> Sent: Wednesday, April 22, 2009 6:00 AM
> To: [e-mail]
> Cc: Anton Borisov; Mondrian Nuessle
> Subject: Re: [coreboot] [PATCH] First support for HP DL145 G3
>
> Hi,
>
> Status of dl145g3:
> 2 Opteron 2218HE's working in the machine. No problem
> both cpu's got 4GB of ram, NUMA is working :)
> booting with filo works
> converted the patches of Mondrian to the last revision of coreboot.
>
> (original patches are on trac, issue 127)
> src/mainboard/hp/dl145g3/Config.lb:
>
> http://merlin.ugent.be/~samuel/dl145g3/config/srcConfig.lb
> src/mainboard/hp/dl145g3/Options.lb:
>
> http://merlin.ugent.be/~samuel/dl145g3/config/srcOptions.lb
> targets/hp/dl145g3/Config.lb:
>
> http://merlin.ugent.be/~samuel/dl145g3/config/Config.lb
> Contacted Anton Borisov to extract the VGA rom of the orginal HP
>
> bios: http://merlin.ugent.be/~samuel/dl145g3/OPROM.00.rom
>
> 2 concerns:
> CPU0 gets very hot during idling. Not really sure what is going on
>
> there. Does the code take into account that this is actually a HE (low
> power opteron)? or is that ignored? How can i check that the voltage
> applied by coreboot is correct? Maybe it's ok.. I just want a
> confirmation on this.
> 2. VGA output is still not working. I have tried both YABEL and
> X86BIOS to load the vga rom but i get errors...
> X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> entering emulator
> int1a vector at 274af
>
> YABEL: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootYABEL.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> pci_cfg_read(): Config read access invalid device! bus: 00 (00),
> devfn: 00 (20), offs: 00
>
> I've also tried with a seperate VGA card on the pci-e slot on the
> machine but that doesn't seem to work either.

Have you tried SeaBIOS for your payload?  It runs the option ROMS directly
instead of emulating them.  As an added bonus it will use your original grub
installation.

If SeaBIOS doesn't work for the onboard graphics or the VGA card, send the
logs.

Thanks,
Myles

[samuel]

> Have you tried SeaBIOS for your payload?  It runs the option ROMS directly
> instead of emulating them.  As an added bonus it will use your original grub
> installation.

I have considered this but didn't test it yet as filo also had a patch
to support the dl145g3, and i'm unsure if i need to port that patch to
seabios as well:
http://www.coreboot.org/pipermail/coreboot/attachments/20090331/9e221f84/att...
> If SeaBIOS doesn't work for the onboard graphics or the VGA card, send the
> logs.
>
> Thanks,
> Myles

[Myles Watson]

On Wed, Apr 22, 2009 at 6:00 AM, samuel  wrote:
> Hi,
>
> Status of dl145g3:
> 2 Opteron 2218HE's working in the machine. No problem
> both cpu's got 4GB of ram, NUMA is working :)
> booting with filo works
> converted the patches of Mondrian to the last revision of coreboot.
>
> (original patches are on trac, issue 127)
>  * src/mainboard/hp/dl145g3/Config.lb:
> http://merlin.ugent.be/~samuel/dl145g3/config/srcConfig.lb
>  * src/mainboard/hp/dl145g3/Options.lb:
> http://merlin.ugent.be/~samuel/dl145g3/config/srcOptions.lb
>  * targets/hp/dl145g3/Config.lb:
> http://merlin.ugent.be/~samuel/dl145g3/config/Config.lb
> Contacted Anton Borisov to extract the VGA rom of the orginal HP
>
> bios: http://merlin.ugent.be/~samuel/dl145g3/OPROM.00.rom
>
> 2 concerns:
> CPU0 gets very hot during idling. Not really sure what is going on
>
> there. Does the code take into account that this is actually a HE (low
> power opteron)? or is that ignored? How can i check that the voltage
> applied by coreboot is correct? Maybe it's ok.. I just want a
> confirmation on this.
> 2. VGA output is still not working. I have tried both YABEL and
> X86BIOS to load the vga rom but i get errors...
> X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> entering emulator
> int1a vector at 274af
>
> YABEL: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootYABEL.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> pci_cfg_read(): Config read access invalid device! bus: 00 (00),
> devfn: 00 (20), offs: 00

device pci 2.2 on end # USB
    				# when HT_CHAIN_END_UNITID_BASE (0,1) < HT_CHAIN_UNITID_BASE (6,,,,),
    				device pci 4.0 on end # VGA  /* REMOVE THIS ONE */

chip drivers/pci/onboard
    					# it is in bcm5785_0 bus, but the device id can
    					# not be changed even unitid is changed, fake one
    					# to get the rom_address
    					# if HT_CHAIN_END_UNITID_BASE=0, it is 4,
    					# if HT_CHAIN_END_UNITID_BASE=1, it is 3
    					device pci 4.0 on end
    					register "rom_address" = "0xfff00000"
    				end

This is probably not _the_ problem, but it is a problem.  Could you
remove the first reference to the VGA.  I think the correct one is
getting disabled.

Thanks,
Myles

[samuel]

> > devfn: 00 (20), offs: 00
>
> device pci 2.2 on end # USB
>                                        # when HT_CHAIN_END_UNITID_BASE (0,1) < HT_CHAIN_UNITID_BASE (6,,,,),
>                                        device pci 4.0 on end # VGA  /* REMOVE THIS ONE */
>
> chip drivers/pci/onboard
>                                                # it is in bcm5785_0 bus, but the device id can
>                                                # not be changed even unitid is changed, fake one
>                                                # to get the rom_address
>                                                # if HT_CHAIN_END_UNITID_BASE=0, it is 4,
>                                                # if HT_CHAIN_END_UNITID_BASE=1, it is 3
>                                                device pci 4.0 on end
>                                                register "rom_address" = "0xfff00000"
>                                        end
>
> This is probably not _the_ problem, but it is a problem.  Could you
> remove the first reference to the VGA.  I think the correct one is
> getting disabled.

removed it and it didn't make a difference :) so at least it was not needed
The error remains the same though

[Tom Sylla]

On Wed, Apr 22, 2009 at 8:00 AM, samuel  wrote:
> VGA output is still not working. I have tried both YABEL and
>
> X86BIOS to load the vga rom but i get errors...
> X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> entering emulator
> int1a vector at 274af
>
> YABEL: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootYABEL.log
> relevant part:
> On mainboard, rom address for PCI: 00:04.0 = fff00000
> copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> pci_cfg_read(): Config read access invalid device! bus: 00 (00),
> devfn: 00 (20), offs: 00

I only see two "int 1a"s in the ROM binary:

seg000:77DC B8 02 B1                          mov     ax, 0B102h
seg000:77DF B9 41 15                          mov     cx, 1541h
seg000:77E2 BA B9 10                          mov     dx, 10B9h
seg000:77E5 BE 00 00                          mov     si, 0
seg000:77E8 CD 1A                             int     1Ah
...
seg000:7801 B8 02 B1                          mov     ax, 0B102h
seg000:7804 B9 43 52                          mov     cx, 5243h
seg000:7807 BA B9 10                          mov     dx, 10B9h
seg000:780A BE 00 00                          mov     si, 0
seg000:780D CD 1A                             int     1Ah

so it is searching for the 0th instance of ven/dev 10b9:1541 or 10b9:5243:
http://www.ctyme.com/intr/rb-2372.htm

Those are:
10b9  ALi Corporation
    1541  M1541
    	10b9 1541  ALI M1541 Aladdin V/V+ AGP System Controller
    5243  M1541 PCI to AGP Controller

which seems kind of weird. Maybe it does some workaround when on a
platform with those devices.

Also, is there any tracing you can enable (or just add) in the
emulators? (to see more closely where it is failing, there is probably
more execution after the int 1a and the config read of 0) I don't know
if YABEL's and X86BIOS's messages are warnings or errors.

[Myles Watson]

On Wed, Apr 22, 2009 at 9:29 AM, Tom Sylla  wrote:
> On Wed, Apr 22, 2009 at 8:00 AM, samuel  wrote:
> > VGA output is still not working. I have tried both YABEL and
> >
> > X86BIOS to load the vga rom but i get errors...
> > X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
> > relevant part:
> > On mainboard, rom address for PCI: 00:04.0 = fff00000
> > copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> > entering emulator
> > int1a vector at 274af
> >
> > YABEL: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootYABEL.log
> > relevant part:
> > On mainboard, rom address for PCI: 00:04.0 = fff00000
> > copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> > pci_cfg_read(): Config read access invalid device! bus: 00 (00),
> > devfn: 00 (20), offs: 00
>
> I only see two "int 1a"s in the ROM binary:
>
> seg000:77DC B8 02 B1                          mov     ax, 0B102h
> seg000:77DF B9 41 15                          mov     cx, 1541h
> seg000:77E2 BA B9 10                          mov     dx, 10B9h
> seg000:77E5 BE 00 00                          mov     si, 0
> seg000:77E8 CD 1A                             int     1Ah
> ...
> seg000:7801 B8 02 B1                          mov     ax, 0B102h
> seg000:7804 B9 43 52                          mov     cx, 5243h
> seg000:7807 BA B9 10                          mov     dx, 10B9h
> seg000:780A BE 00 00                          mov     si, 0
> seg000:780D CD 1A                             int     1Ah
>
> so it is searching for the 0th instance of ven/dev 10b9:1541 or 10b9:5243:
> http://www.ctyme.com/intr/rb-2372.htm
>
> Those are:
> 10b9  ALi Corporation
>        1541  M1541
>                10b9 1541  ALI M1541 Aladdin V/V+ AGP System Controller
>        5243  M1541 PCI to AGP Controller
>
> which seems kind of weird. Maybe it does some workaround when on a
> platform with those devices.
>
> Also, is there any tracing you can enable (or just add) in the
> emulators? (to see more closely where it is failing, there is probably
> more execution after the int 1a and the config read of 0) I don't know
> if YABEL's and X86BIOS's messages are warnings or errors.

It turns out that this was caused by an extra VGA entry in the Config.lb

# when HT_CHAIN_END_UNITID_BASE
(0,1) < HT_CHAIN_UNITID_BASE (6,,,,),
                                       device pci 4.0 on end # VGA  /*
REMOVE THIS ONE */

chip drivers/pci/onboard
                                               # it is in bcm5785_0
bus, but the device id can
                                               # not be changed even
unitid is changed, fake one
                                               # to get the rom_address
                                               # if
HT_CHAIN_END_UNITID_BASE=0, it is 4,
                                               # if
HT_CHAIN_END_UNITID_BASE=1, it is 3
                                               device pci 4.0 on end
                                               register "rom_address"
= "0xfff00000"
                                       end

It was trying to initialize the device, but it wasn't correctly
connected to the bus.  Here's the new failure.  Now it looks like it's
not finding the ROM anymore.

new log: http://merlin.ugent.be/~samuel/dl145g3/vgarom/coreboot-3.log

Thanks,
Myles

[Myles Watson]

On Wed, Apr 22, 2009 at 10:59 AM, Myles Watson  wrote:
> On Wed, Apr 22, 2009 at 9:29 AM, Tom Sylla  wrote:
> > On Wed, Apr 22, 2009 at 8:00 AM, samuel  wrote:
> > > VGA output is still not working. I have tried both YABEL and
> > >
> > > X86BIOS to load the vga rom but i get errors...
> > > X86BIOS: http://merlin.ugent.be/~samuel/dl145g3/vgarom/corebootx86bios.log
> > > relevant part:
> > > On mainboard, rom address for PCI: 00:04.0 = fff00000
> > > copying VGA ROM Image from fff00000 to 0xc0000, 0x8000 bytes
> > > entering emulator
> > > int1a vector at 274af

from the latest log:

PCI: 00:04.0 init
On card, rom address for PCI: 00:04.0 = fda00000
Incorrect Expansion ROM Header Signature 0000

Samuel,

It should still say On mainboard.  Could you send the snippet from the
Config.lb that you're using?

Thanks,
Myles

[Paul Menzel]

Am Donnerstag, den 02.04.2009, 19:12 +0200 schrieb samuel:
> OK,
>
> I managed booting the firmware... I never worked with serial ports and
> stuff so this took me a while to grab the idea...
>
> Anyway,
> you can find the boot.log on http://merlin.ugent.be/~samuel/dl145g3/boot.log
>
> Obviously that boot line is wrong.
> http://merlin.ugent.be/~samuel/dl145g3/filo_config
> obviously CONFIG_MENULST_FILE="hda3:/boot/filo/menu.lst" is wrong. My
> kernel I'd like to boot is hda1:/2.6.28.8_x86-64
>
> Trying to boot this as follows:
>
> kernel hda1:/2.6.28.8_x86_64
>
> results in:
>
> Drive 0 does not exist
> Error 15: File not found.
>
> I'm a bit stuck I'm afraid, afaik i did patch filo with the patch from
> mondrian so prolly me config of filo is wrong?

I have never tried it myself. But as far as I know FILO handles config
files from GRUB, where is written what options are available for boot.
So if you have used GRUB before, you have to point the config to this
file.

Thanks,

Paul

[Myles Watson]

2009/3/31 Mondrian Nuessle :
> Hi everybody,
>
> I prepared an initial set of patches to support the HP DL145 G3 server, which is a Broadcom HT-1000/HT-2100 based Socket
> F machine.

Thanks for the patches!  Could you add a Signed-off-by: line so I can
ack and commit it?

Thanks,
Myles

[samuel]

Acked-by: Samuel Verstraete <samuel.verstraete at gmail.com>

2009/3/31 Mondrian Nuessle :
> Hi everybody,
>
> I prepared an initial set of patches to support the HP DL145 G3 server, which is a Broadcom HT-1000/HT-2100 based Socket
> F machine.
>
> The current status of the port is:
> we have serial output (using the serverengines pilot chip on the board, see attached pilot_superio.patch)
> CPU and memory gets initialized
> on-board eth0 is working
> SATA is working
> the on-board graphic was not tested (not important for
>
> a server and the graphics sucks anyway...)
> USB and PS2 work
> no ACPI
> I remember that I could also use a PCIe card, see attached bcm21000.patch
> HTX card does work.
> Earlier problems we had in respect to timer interrupt are not there anymore.
> the hp_dl145_g3.patch includes the mainboard and the target dirs
>
> BUT:
> if I coldboot, the CPUs are only running at 1 GHz and not with 2.6 GHz as they should...
> if we boot the factory BIOS and then perform a warm-reset into Coreboot, the CPUs run with 2.6 GHz.
>
> I hope someone on the list can help.
> There are also some more problems with the HT-1000 (aka bcm5785) southbridge:
>
> - if I activate IDE in bcm5785_early_setup.c, the machine will freeze shortly after being completely booted
>  - if I activate the second IOAPIC (line 179 in bcm5785_early_setup.c), my linux boot hangs when starting udevd...
>  - with both features disabled (see attached bcm5785_noide.patch), the machine boots and does not crash, of course
>    PATA drives are not available.
>
> To boot the machine, I tested FILO. The on-board SATA adapter uses a class id of 0x104 in sata mode
> (which we use). FILO does not accept this class ID; I prepared a patch that adds this class ID to
> the respective if statement in driver/ide.c. FILO does then boot from SATA without any problems (see attached
> filo_dl145_sata.patch).
>
> I hope this gets you started Samuel, and I hope the list can provide some insight regarding the CPU
> frequency problem.
>
> Regards,
> Mondrian

[Myles Watson]

On Wed, Apr 22, 2009 at 2:22 PM, samuel  wrote:
> Acked-by: Samuel Verstraete <samuel.verstraete at gmail.com>

Rev 4182, 4183, 4184.

I didn't commit the patch that disables IDE since it would affect
other boards.  It will need to be reworked so that it is
board-specific.

Thanks,
Myles