Hello,

recently, a kernel patch series was posted to fix issues with HugeTLB
on UltraSPARC (sun4u) machines [1]. I have built a test Debian kernel
which includes the patch and uploaded it to [2].

Could someone test this kernel and report back whether it causes any regressions and also whether it possible fixes the stability issues
we're seeing on UltraSPARC machines with newer kernels?

Thanks,
Adrian

[1] https://marc.info/?l=linux-sparc&m=175262905900358&w=2
[2] https://people.debian.org/~glaubitz/linux-image-6.12.38+1-sparc64-smp_6.12.38-1+sparc64_sparc64.deb

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Marcelo,

On Thu, 2025-07-17 at 15:00 +0200, Marcelo Bezerra wrote:

Is there an easy way to get an installer iso with this kernel?

I have access to both a blade 150 a ultra 45 I can use to test this.

In principle that would be possible, but it would cost me some time to
build such an image as the patched kernel is not part of the official distribution yet.

However, you can just install from an older image such as [1] and
manually install the kernel there.

If it's absolutely not possible for anyone to test this, I can build
an ISO image later this week.

Adrian

[1] https://cdimage.debian.org/cdimage/ports/snapshots/2019-07-16/debian-10.0-sparc64-NETINST-1.iso

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Marcelo,

On Thu, 2025-07-17 at 18:45 +0200, Marcelo Bezerra wrote:

I managed to re-install debian on the ultra45, but I get a panic on
both the latest 6.12.38 or your kernel. Probably unrelated to your
changes though. this has been an issue on this ultra 45 for a while
now.

The stack traces were different, but they both panicked due to a
corrupted stack inside the scheduler.

Thanks a lot for testing, much appreciated!

Please keep your machine available, I might follow up with another
kernel shortly.

Adrian

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

On Thu, 2025-07-17 at 08:53 -0600, Stan Johnson wrote:

On a Sparc Ultra-30, booting stops at "Fast Data Access MMU Miss".

And booting the standard 6.12.38 kernel [1] without the patch works?

Adrian

[1] http://ftp.ports.debian.org/debian-ports/pool-sparc64/main/l/linux/linux-image-6.12.38+deb13-sparc64_6.12.38-1_sparc64.deb

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

HI Adrian,

John Paul Adrian Glaubitz wrote:

Hello,

recently, a kernel patch series was posted to fix issues with HugeTLB
on UltraSPARC (sun4u) machines [1]. I have built a test Debian kernel
which includes the patch and uploaded it to [2].

Could someone test this kernel and report back whether it causes any regressions and also whether it possible fixes the stability issues
we're seeing on UltraSPARC machines with newer kernels?

I gave the kernel a spin on my T2000 with Niagara. Sorry, took a while,
but needed a spare evening, possibility to shut down, connect LIOM cable
to another system, ssh.. etc etc.

I left it running a couple of hours, did run git on big repositories
several time, run aptitude, compiled stuff on 32 cores. Exported GUI
stuff thorugh X11.

No issues noted, no regressions at a first test.

Furthermore, previous kernel usually crashed itself on a hot reboot,
requiring always a cold start (this caused me a lot of trouble testing
in the past). I cycled a couple of reboots without issues.

No certainty, but for me this kernel is better than the previous one!

I did not run explicit stress test. I remember there was one floating
around on this list that did lock up my system after 10 minutes or so.
would be interesting to try.

Riccardo

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Riccardo,

On Wed, 2025-07-23 at 02:26 +0200, Riccardo Mottola wrote:

I gave the kernel a spin on my T2000 with Niagara. Sorry, took a while,
but needed a spare evening, possibility to shut down, connect LIOM cable
to another system, ssh.. etc etc.

Thanks for testing. However, the T2000 is unfortunately not relevant for
this test as the bug affects sun4u machines only while the T2000 is sun4v.


We need test on machines older than the T1000 which was the first sun4v machine.

Adrian

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi!

John Paul Adrian Glaubitz wrote:

Thanks for testing. However, the T2000 is unfortunately not relevant for
this test as the bug affects sun4u machines only while the T2000 is sun4v.

better a test more than one less, at least no regression!
Maybe the improvement comes from other fixes since I was running an
slightly older kernel anyway. Good to know.

We need test on machines older than the T1000 which was the first sun4v machine.

Can't help with that at the moment, albeit I will probably convert
certain systems from Solaris to BSD and Linux in the future.

Riccardo

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi,

On Wed, 2025-07-23 at 10:51 +0200, Riccardo Mottola wrote:

John Paul Adrian Glaubitz wrote:

Thanks for testing. However, the T2000 is unfortunately not relevant for this test as the bug affects sun4u machines only while the T2000 is sun4v.

better a test more than one less, at least no regression!

Sure. But the patch doesn't touch any sun4v code, so I wouldn't expect any regressions there.

Maybe the improvement comes from other fixes since I was running an
slightly older kernel anyway. Good to know.

Could be.

We need test on machines older than the T1000 which was the first sun4v machine.

Can't help with that at the moment, albeit I will probably convert
certain systems from Solaris to BSD and Linux in the future.

Do you have any sun4u machines?

Adrian

--
.''`. John Paul Adrian Glaubitz
: :' : Debian Developer
`. `' Physicist
`- GPG: 62FF 8A75 84E0 2956 9546 0006 7426 3B37 F5B5 F913

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

Hi Adrian!

John Paul Adrian Glaubitz wrote:

Can't help with that at the moment, albeit I will probably convert
certain systems from Solaris to BSD and Linux in the future.

Do you have any sun4u machines?

yes I do an aging Ultra 1 should be the definition of sun4u  and a Netra
T1 105 should also qualify, doesn't it?


Right now however, they are running old Solaris, I cannot test  your
kernel easily, although I could boot off cd-rom.
Or maybe I find a spare HD with sled and do an install

Riccardo

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

I'd be willing to test on my Ultra1 but I haven't been able to get debian installed. If there was an install ISO with this kernel I can test it.


On Wed, Jul 23, 2025 at 6:55 AM Riccardo Mottola <riccardo.mottola@libero.it> wrote:

Hi Adrian!

John Paul Adrian Glaubitz wrote:

Can't help with that at the moment, albeit I will probably convert
certain systems from Solaris to BSD and Linux in the future.

Do you have any sun4u machines?

yes I do an aging Ultra 1 should be the definition of sun4u and a Netra
T1 105 should also qualify, doesn't it?

Right now however, they are running old Solaris, I cannot test your
kernel easily, although I could boot off cd-rom.
Or maybe I find a spare HD with sled and do an install

Riccardo

--
Jeremy Leonard
JeremyL@elite4god.com
Cell: (517) 285-8309

<div dir="ltr">I&#39;d be willing to test on my Ultra1 but I haven&#39;t been able to get debian installed. If there was an install ISO with this kernel I can test it.<div><br></div></div><br><div class="gmail_quote gmail_quote_container"><div dir="ltr"
class="gmail_attr">On Wed, Jul 23, 2025 at 6:55 AM Riccardo Mottola &lt;<a href="mailto:riccardo.mottola@libero.it">riccardo.mottola@libero.it</a>&gt; wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid
rgb(204,204,204);padding-left:1ex">Hi Adrian!<br>

John Paul Adrian Glaubitz wrote:<br>
&gt;&gt; Can&#39;t help with that at the moment, albeit I will probably convert<br>
&gt;&gt; certain systems from Solaris to BSD and Linux in the future.<br>
&gt; Do you have any sun4u machines?<br>

yes I do an aging Ultra 1 should be the definition of sun4u  and a Netra <br> T1 105 should also qualify, doesn&#39;t it?<br>

On 7/23/2025 10:40 AM, Jeremy Leonard wrote:

I'd be willing to test on my Ultra1 but I haven't been able to get
debian installed. If there was an install ISO with this kernel I can
test it.

On Wed, Jul 23, 2025 at 6:55 AM Riccardo Mottola <riccardo.mottola@libero.it <mailto:riccardo.mottola@libero.it>> wrote:

Hi Adrian!

John Paul Adrian Glaubitz wrote:
>> Can't help with that at the moment, albeit I will probably convert
>> certain systems from Solaris to BSD and Linux in the future.
> Do you have any sun4u machines?

yes I do an aging Ultra 1 should be the definition of sun4u  and a
Netra
T1 105 should also qualify, doesn't it?

Right now however, they are running old Solaris, I cannot test  your
kernel easily, although I could boot off cd-rom.
Or maybe I find a spare HD with sled and do an install

Riccardo

--
Jeremy Leonard
JeremyL@elite4god.com <mailto:JeremyL@elite4god.com>
Cell: (517) 285-8309

Hello All,

Sorry I'm a bit late to the conversation on this topic (I've been
distracted working on an old RS/6000 for the last week)... Is there
still a need for testing on sun4u for this? I have a crash-and-burn
Ultra5 running for just this use case if it would be helpful.

Cheers,
Bill

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)