Alpha hardware for printk tests?

Alpha hardware for printk tests?

[John Ogness]

Hello,

I have posted[0] a new lockless printk ringbuffer implementation. The
ringbuffer has been tested heavily on x86_64 and arm64 SMP
systems. However, I would also like to test it on an SMP Alpha
system. Since the main purpose of the test is to verify the memory
barriers, the tests need to run on real hardware.

The new ringbuffer (with test routine) can also be built as a kernel
module and does not require any kernel patching. It should work on any
Linux kernel 4.20 or higher.

Is there anyone with access to SMP Alpha hardware that would be willing
to run the tests for me? Or can someone tell me where I might find such
a candidate? Thanks!

John Ogness

[0] https://lkml.kernel.org/r/20200501094010.17694-1-john.ogness@linutronix.de

Re: Alpha hardware for printk tests?

[Michael Cree]

On Tue, May 12, 2020 at 04:39:06PM +0200, John Ogness wrote:
> I have posted[0] a new lockless printk ringbuffer implementation. The
> ringbuffer has been tested heavily on x86_64 and arm64 SMP
> systems. However, I would also like to test it on an SMP Alpha
> system. Since the main purpose of the test is to verify the memory
> barriers, the tests need to run on real hardware.
> 
> The new ringbuffer (with test routine) can also be built as a kernel
> module and does not require any kernel patching. It should work on any
> Linux kernel 4.20 or higher.
> 
> Is there anyone with access to SMP Alpha hardware that would be willing
> to run the tests for me? Or can someone tell me where I might find such
> a candidate? Thanks!

I run the two Alpha ES45's building for Debian-Ports.  I can run
the tests on one of them if you provide me with instructions.

Cheers,
Michael.

Re: Alpha hardware for printk tests?

[John Ogness]

On 2020-05-13, Michael Cree <mcree@orcon.net.nz> wrote:
>> I have posted[0] a new lockless printk ringbuffer implementation. The
>> ringbuffer has been tested heavily on x86_64 and arm64 SMP
>> systems. However, I would also like to test it on an SMP Alpha
>> system. Since the main purpose of the test is to verify the memory
>> barriers, the tests need to run on real hardware.
>> 
>> The new ringbuffer (with test routine) can also be built as a kernel
>> module and does not require any kernel patching. It should work on
>> any Linux kernel 4.20 or higher.
>> 
>> Is there anyone with access to SMP Alpha hardware that would be
>> willing to run the tests for me? Or can someone tell me where I might
>> find such a candidate? Thanks!
>
> I run the two Alpha ES45's building for Debian-Ports.  I can run
> the tests on one of them if you provide me with instructions.

That would be very helpful! Actually the test should build/run fine on
the Debian/stable (4.19) kernel as well. The source code is available
here:

https://linux:linux@dl.linutronix.de/customers/printk/prb-test.tar.gz

The README in the tarball explains how to build/run the test and what
information to send back after the test.

Even if you only had time to run the test for a few minutes, it would be
very helpful for me!

If you have any problems with the test module, please let me know. This
test has been confirmed to run successfully on: arm64, amd64, ppc64el

John Ogness

Re: Alpha hardware for printk tests?

[John Ogness]

On 2020-05-12, John Ogness <john.ogness@linutronix.de> wrote:
> I have posted[0] a new lockless printk ringbuffer implementation. The
> ringbuffer has been tested heavily on x86_64 and arm64 SMP
> systems. However, I would also like to test it on an SMP Alpha
> system. Since the main purpose of the test is to verify the memory
> barriers, the tests need to run on real hardware.
> 
> The new ringbuffer (with test routine) can also be built as a kernel
> module and does not require any kernel patching. It should work on any
> Linux kernel 4.20 or higher.
> 
> Is there anyone with access to SMP Alpha hardware that would be
> willing to run the tests for me? Or can someone tell me where I might
> find such a candidate? Thanks!

A problem leading to data corruption in the ringbuffer messages has been
identified and fixed. A new version of the test module is available:

https://linux:linux@dl.linutronix.de/customers/printk/prb-test.tar.gz

The README in the tarball explains how to build/run the test and what
information to send back after the test.

This test module should build/run without any kernel patching or
rebooting. Even if it is only allowed to run on your SMP-Alpha for an
hour, that would be very helpful for me.

Thanks,
John Ogness

Re: Alpha hardware for printk tests?

[Michael Cree]

On Tue, May 19, 2020 at 03:42:15PM +0200, John Ogness wrote:
> On 2020-05-12, John Ogness <john.ogness@linutronix.de> wrote:
> > I have posted[0] a new lockless printk ringbuffer implementation. The
> > ringbuffer has been tested heavily on x86_64 and arm64 SMP
> > systems. However, I would also like to test it on an SMP Alpha
> > system. Since the main purpose of the test is to verify the memory
> > barriers, the tests need to run on real hardware.
> > 
> > The new ringbuffer (with test routine) can also be built as a kernel
> > module and does not require any kernel patching. It should work on any
> > Linux kernel 4.20 or higher.
> > 
> > Is there anyone with access to SMP Alpha hardware that would be
> > willing to run the tests for me? Or can someone tell me where I might
> > find such a candidate? Thanks!
> 
> A problem leading to data corruption in the ringbuffer messages has been
> identified and fixed. A new version of the test module is available:
> 
> https://linux:linux@dl.linutronix.de/customers/printk/prb-test.tar.gz
> 
> The README in the tarball explains how to build/run the test and what
> information to send back after the test.
> 
> This test module should build/run without any kernel patching or
> rebooting. Even if it is only allowed to run on your SMP-Alpha for an
> hour, that would be very helpful for me.

Thanks for that.  I will give this new version a test.  I hadn't run
the tests as yet because the two alphas were very busy building
packages, but things are quiet again, so I'll take one out of the
builder network for testing.

Cheers,
Michael.

Re: Alpha hardware for printk tests?

[Michael Cree]

Hi John,

On Tue, May 19, 2020 at 03:42:15PM +0200, John Ogness wrote:
> On 2020-05-12, John Ogness <john.ogness@linutronix.de> wrote:
> > I have posted[0] a new lockless printk ringbuffer implementation. The
> > ringbuffer has been tested heavily on x86_64 and arm64 SMP
> > systems. However, I would also like to test it on an SMP Alpha
> > system. Since the main purpose of the test is to verify the memory
> > barriers, the tests need to run on real hardware.

I ran it for about 48 hours on an ES45 with two EV68CB CPUs.

This is the dmesg output.  Nothing untoward shows up.
The unaligned traps are only from some builds of R software that
I allowed to happen at some point during the 48 hours.

[5594108.751049] prb: loading out-of-tree module taints kernel.
[5594108.752025] prbtest: starting test
[5594108.795971] prbtest: start thread 000 (writer)
[5594108.795971] prbtest: start thread 001 (reader)
[5633065.742821] do_entUnaUser: 267 callbacks suppressed
[5633065.742821] R(31934): unaligned trap at 00000200034f64e4: 000000004dc6b4c5 2c 1
[5633065.899071] R(31934): unaligned trap at 00000200034f64ec: 00000000688e1c08 2c 1
[5633065.995751] R(31934): unaligned trap at 00000200034f64f4: 000000007ab1d76f 2c 1
[5633066.093407] R(31934): unaligned trap at 00000200034f64fc: 0000000079eeaefd 2c 1
[5633066.194970] R(31934): unaligned trap at 00000200034f64e4: 000000004dc6b4c5 2c 1
[5633802.658460] do_entUnaUser: 399 callbacks suppressed
[5633802.658460] R(5173): unaligned trap at 00000200034f64e4: 00000000bc6e5800 2c 1
[5633802.813733] R(5173): unaligned trap at 00000200034f64ec: 0000000099c12496 2c 1
[5633802.908460] R(5173): unaligned trap at 00000200034f64f4: 00000000664dc5fb 2c 1
[5633803.005139] R(5173): unaligned trap at 00000200034f64fc: 00000000ac4dd9c5 2c 1
[5633803.104749] R(5173): unaligned trap at 00000200034f64e4: 00000000bc6e5800 2c 1
[5765821.511765] reader1: total_lost=1320184974 max_lost=66240 total_read=47352899655 seq=48696527777
[5765821.619186] prbtest: end thread 001 (reader)
[5765821.671921] prbtest: end thread 000 (wrote 48696598157, max/avg/min 1953124/2578/0)
[5765822.735397] prbtest: completed test

There is no /sys/kernel/debug/tracing directory on this machine.
Maybe I need to enable some kernel config in the kernel build?

Cheers,
Michael.