sparc32 2.6 work

sparc32 2.6 work

[Heitzso]

(if there's a specific email list I should focus on please let me know,
or a particular individual/mentor I can work through on this to avoid
having simple/trivial emails hassle a list .... thanks)

I figured out patching 2.6.12 rc against base 2.6.11 and not 2.6.11.X

I setup serial console as recommended by a few folks to better
monitor the boot sequence.

Next minor problem was/is that modules were not automagically loading
(though that was set in .config) so every module that may be
necessary has to be explicitly tagged in
/etc/modules.autoload.d/kernel-2.6
This may be gentoo specific, I don't know.  But I needed to
add in the HME/LANCE modules so the network would crank up
and sshd start, allowing me to sshd in (as well as serial console
login), along with others.

With that working I reattached keyboard/monitor and rebooted.
Still getting initial invisible boot screen, but by the time
the gentoo /etc/init.d scripts are running the monitor is
visible.

Trying to log on with the attached keyboard/monitor is rather
horribly broken in the keyboard.  I have sun keyboard compiled
into the kernel (not a module).  Don't have a competing keyboard
compiled in.  My config was based on 2.4.30 config run through
oldconfig, then hand tweaked.  (and my 2.4.30 SMP gives good
messages to attached monitor/keyboard, keyboard works, and
X works for at least a few minutes before locking up the 2.4.30
SMP kernel).

Then, running 2.6.12-rc4 UP on ross hypersparc ss20, after being
up about 8 hours and while editing in vi, the filesystem(s) just
disappeared.  By disappearing "ls" returns blank and "df -m"
hangs.

I have a regular single supersparcII cpu that I could drop into
the ss20 if that's recommended, but the whole gentoo env is
compiled with some ross hypersparc flags.  I don't know if that's
a problem or not.

SO ....

I've offered to try and help with the sparc32 port of 2.6 and
am wondering where I should start.  Seems like it makes sense
to get UP solid before jumping into the world of SMP, but it
may be my above problems are not really problems, or have simple
solutions.

If someone could do a little hand holding and/or point out a
small area of code for me to jump into I'll be glad to try to
help fix the code.  But I need that mentoring because the above
problems, much less SMP, seem to be all over the map ...


Thanks,
Heitzso

Re: sparc32 2.6 work

[Jurij Smakov]

Hi Heitzso,

On Sat, 14 May 2005, Heitzso wrote:

> (if there's a specific email list I should focus on please let me know,
> or a particular individual/mentor I can work through on this to avoid
> having simple/trivial emails hassle a list .... thanks)

Please address all the user-level questions to the support channels of the 
distribution of your choice.

> I've offered to try and help with the sparc32 port of 2.6 and
> am wondering where I should start.  Seems like it makes sense
> to get UP solid before jumping into the world of SMP, but it
> may be my above problems are not really problems, or have simple
> solutions.

Well, UP in Debian is pretty solid on sparc32. I happen to own one of the 
machines with Hypersparc CPU and UP 2.6.x and Debian's pre-built kernels 
are stable enough to build other packages and kernels on it. The only 
sparc32 patch we are applying against 2.6.11 is the following:

--- kernel-source-2.6.11-2.6.11-orig//arch/sparc/mm/srmmu.c	2005-03-02 08:38:10.000000000 +0100
+++ kernel-source-2.6.11-2.6.11/arch/sparc/mm/srmmu.c	2005-03-05 10:39:38.677679448 +0100
@@ -1464,7 +1464,6 @@
  static void __init init_hypersparc(void)
  {
          srmmu_name = "ROSS HyperSparc";
-	srmmu_modtype = HyperSparc;

          init_vac_layout();

Without it the kernel fails to mount initrd on boot due to unknown reason. 
This change was discovered after numerous experiments :-). The only other 
quirk I know (which can be related) is that Debian's 2.6.8 netboot images 
do not boot Hypersparc machines (others seem to be fine), again due to the 
trouble with initrd.

> If someone could do a little hand holding and/or point out a
> small area of code for me to jump into I'll be glad to try to
> help fix the code.  But I need that mentoring because the above
> problems, much less SMP, seem to be all over the map ...

debian-sparc mailing list has been fairly active recently. I am sure 
people there would be glad to help you with your issues (given that you 
will be willing to <subliminal message>use Debian</subliminal message>, of 
course).

>
> Thanks,
> Heitzso

Best regards,

Jurij Smakov                                        jurij@wooyd.org
Key: http://www.wooyd.org/pgpkey/                   KeyID: C99E03CC

Re: sparc32 2.6 work

[Bob Breuer]

Jurij Smakov wrote:
> 
> Well, UP in Debian is pretty solid on sparc32. I happen to own one of 
> the machines with Hypersparc CPU and UP 2.6.x and Debian's pre-built 
> kernels are stable enough to build other packages and kernels on it. The 
> only sparc32 patch we are applying against 2.6.11 is the following:
> 
> --- kernel-source-2.6.11-2.6.11-orig//arch/sparc/mm/srmmu.c    
> 2005-03-02 08:38:10.000000000 +0100
> +++ kernel-source-2.6.11-2.6.11/arch/sparc/mm/srmmu.c    2005-03-05 
> 10:39:38.677679448 +0100
> @@ -1464,7 +1464,6 @@
>  static void __init init_hypersparc(void)
>  {
>          srmmu_name = "ROSS HyperSparc";
> -    srmmu_modtype = HyperSparc;
> 
>          init_vac_layout();
> 
> Without it the kernel fails to mount initrd on boot due to unknown 
> reason. This change was discovered after numerous experiments :-). The 
> only other quirk I know (which can be related) is that Debian's 2.6.8 
> netboot images do not boot Hypersparc machines (others seem to be fine), 
> again due to the trouble with initrd.
> 

Reverting that change should not be needed for 2.6.11 (but it did cause 
other problems with 2.6.10).  Mounting an initrd under 2.6.11 works for 
me on my hypersparc ss20.  I think there was a memcpy fix for the initrd 
failures.  Anyway, if there is still a problem that requires the change, 
I'm willing to help track it down.

Bob

Re: sparc32 2.6 work

[Jurij Smakov]

On Mon, 16 May 2005, Bob Breuer wrote:
>
> Reverting that change should not be needed for 2.6.11 (but it did cause other 
> problems with 2.6.10).  Mounting an initrd under 2.6.11 works for me on my 
> hypersparc ss20.  I think there was a memcpy fix for the initrd failures. 
> Anyway, if there is still a problem that requires the change, I'm willing to 
> help track it down.
>
> Bob

Hi Bob,

I have now tried building Debian's 2.6.11 kernel with and without this 
change applied, however in both cases it fails to mount the initrd. I'll 
build the vanilla kernel without any extra patches tomorrow and check how 
it will behave.

Thanks and best regards,

Jurij Smakov                                        jurij@wooyd.org
Key: http://www.wooyd.org/pgpkey/                   KeyID: C99E03CC