K8 SMP broken?

[Torsten Duwe]

Hi all,

has anybody tested K8 SMP recently? I've built r5689 for
a Tyan Thunder K8W (s2885, dual socket 940), but it hangs.
Commenting out allow_all_aps_stop(bsp_apicid) brings it up,
but with a single CPU.

Anyone feeling guilty? ;-)

Torsten

[Myles Watson]

> has anybody tested K8 SMP recently?

In this thread:
Re: [coreboot] DL145 G1 with dual dualcore CPU using coreboot ?

Oskar is testing K8 SMP.  It seems to work for him.  He's basing his port
off the s2881.  Maybe you could compare the s2881 and s2885 code.

Thanks,
Myles

[Torsten Duwe]

On Thursday 12 August 2010, Myles Watson wrote:
> > has anybody tested K8 SMP recently?
>
> In this thread:
> Re: [coreboot] DL145 G1 with dual dualcore CPU using coreboot ?
>
> Oskar is testing K8 SMP.  It seems to work for him.

Thank you Myles, good point.
> He's basing his port
> off the s2881.  Maybe you could compare the s2881 and s2885 code.

I have reordered romstage.c to match the 2881; still no go. Maybe something 
with the HT setup.

Torsten

[Myles Watson]

On Fri, Aug 13, 2010 at 2:20 AM, Torsten Duwe  wrote:
> On Thursday 12 August 2010, Myles Watson wrote:
> > > has anybody tested K8 SMP recently?
> >
> > In this thread:
> > Re: [coreboot] DL145 G1 with dual dualcore CPU using coreboot ?
> >
> > Oskar is testing K8 SMP.  It seems to work for him.
>
> Thank you Myles, good point.
> > He's basing his port
> > off the s2881.  Maybe you could compare the s2881 and s2885 code.
>
> I have reordered romstage.c to match the 2881; still no go. Maybe something
> with the HT setup.

You could send the early log to see if we can spot something.

I just tried SimNOW (serengeti_cheetah), and it works with two
single-core CPUs.  The only thing I thought of was maybe you should
change CONFIG_APIC_ID_OFFSET to 0.  It seems like the problem may be
ID confusion if it's trying to stop something that isn't responding?
I'm not sure at all, but I don't think it will hurt to try.

Good luck,
Myles

openSUSE-11.3 gcc broken (Was: K8 SMP broken?)

[Torsten Duwe]

Stop the press! False alarm.
make CC="gcc-4.3 -m32" produces a working image on my openSUSE 11.3,
where the default compiler failed as described. Sorry for the noise.

Torsten

[Stefan Reinauer]

On 8/16/10 5:06 PM, Torsten Duwe wrote:
> Stop the press! False alarm.
> make CC="gcc-4.3 -m32" produces a working image on my openSUSE 11.3,
> where the default compiler failed as described. Sorry for the noise.

Can you please verify whether the compiler built by util/crossgcc
creates a working image for you?

[Rudolf Marek]

> Anyone feeling guilty? ;-)

Is it pre fam10h? It was always big mystery for me how it can work ;) The AP got 
the SAME stack... If you don't believe check for yourself. I think simple fix 
would be to do same thing as FAM10h code is doing.

Please can you try attached patch? Dunno if it works looks like it could.

Thanks,
Rudolf

[Torsten Duwe]

On Thursday 12 August 2010 21:54:49 Rudolf Marek wrote:
> Is it pre fam10h? It was always big mystery for me how it can work ;) The
> AP got the SAME stack... If you don't believe check for yourself.

Oskar seems to be succesful with the current code. Oskar, do you have 
the old 248s running or did you upgrade to the 280s first?
> I think
> simple fix would be to do same thing as FAM10h code is doing.
>
> Please can you try attached patch? Dunno if it works looks like it could.

No, it doesn't.

| coreboot-4.0-r Fri Aug 13 11:45:56 CEST 2010 starting...
| Enabling routing table for node 00 done.
| Enabling SMP settings
| (0,1) link=01
| (1,0) link=01
| setup_remote_node: done
| Renaming current temporary node to 01 done.
| Enabling routing table for node 01 done.
| 02 nodes iniÿ
| 
| INIT detected from  --- { APICID = 00 NODEID = 00 COREID = 00} ---
| 
| Issuing SOFT_RESET...

This is repeated over and over.

Torsten

[Oskar Enoksson]

Torsten Duwe wrote:
> On Thursday 12 August 2010 21:54:49 Rudolf Marek wrote:
> > Is it pre fam10h? It was always big mystery for me how it can work ;) The
> > AP got the SAME stack... If you don't believe check for yourself.
>
> Oskar seems to be succesful with the current code. Oskar, do you have 
> the old 248s running or did you upgrade to the 280s first?

I tried to put my latest flashimage into a dual 248 server, and I had
the same behaviour that Rudolf described: an infinite loop or resets.

With some DIMM configurations I have seen the same behaviour with the
dual 280 also. Perhaps it's a memory configuration problem.

[Mick Reed]

> Torsten, what CPUs do you have?  I have a s2885 with 290s, I can easily* test your rom.

Sorry, I can't test it as easily* as I thought - nevermind.