BETA edition rcu errors

[8bit]

I ran the test and got fails inmediately: https://pastebin.com/xbTeksJN

The second ran get the fails inmediately too.

Looks like the errors I was getting. Are you running the latest kernel? I just upgraded and they released it 3 days ago.

[some_pinephone_user]

Yes. I'm running 5.14.15-1. Anyway, I ran the test on 5.14.14-2 too with same results.

[8bit]

Yes. I'm running 5.14.15-1. Anyway, I ran the test on 5.14.14-2 too with same results.

Hopefully @smaeul will see this and take a look at your errors.

Is this a new PinePhone, has it only started doing this?

[some_pinephone_user]

Yes. I'm running 5.14.15-1. Anyway, I ran the test on 5.14.14-2 too with same results.

Hopefully @smaeul will see this and take a look at your errors.

Is this a new PinePhone, has it only started doing this?

It is a new one.

[some_pinephone_user]

Any news? How the errors should look like? When "skip" is not 0?

[some_pinephone_user]

I did some new test. The RCU failre on deischarging battery logging battery levels. The results are pretty strange: https://dropmefiles.com/4IRLI
Sometimes the test-cycle completes without any error and sometimes with tens of thousand.

@smaeul , could you take a look into?

[some_pinephone_user]

I  could manage it disabling sustemd-timesyncd and installing chrony. On Manjaro ARM it looks like this

sudo systemctl mask systemd-timesyncd
sudo systemctl stop systemd-timesyncd
sudo pacman -S chrony

[smaeul]

I did some new test. The RCU failre on deischarging battery logging battery levels. The results are pretty strange: https://dropmefiles.com/4IRLI
Sometimes the test-cycle completes without any error and sometimes with tens of thousand.

@smaeul , could you take a look into?

Your logs show time jumps that are smaller than those seen on any device before. They are not covered by the existing workaround in the kernel, so an additional kernel change is needed to resolve the errors you are seeing.

The quickest way to fix your time jumps is to apply this patch again, but instead of changing the constant from a 9 to an 8, change it from an 8 to a 7:

https://lore.kernel.org/linux-sunxi/2021...lland.org/

Decreasing the mask size could have a slight performance impact, so there has been some discussion about finding a better way to filter out the jumps.

[some_pinephone_user]

I did some new test. The RCU failre on deischarging battery logging battery levels. The results are pretty strange: https://dropmefiles.com/4IRLI
Sometimes the test-cycle completes without any error and sometimes with tens of thousand.

@smaeul , could you take a look into?

Your logs show time jumps that are smaller than those seen on any device before. They are not covered by the existing workaround in the kernel, so an additional kernel change is needed to resolve the errors you are seeing.

The quickest way to fix your time jumps is to apply this patch again, but instead of changing the constant from a 9 to an 8, change it from an 8 to a 7:

https://lore.kernel.org/linux-sunxi/2021...lland.org/

Decreasing the mask size could have a slight performance impact, so there has been some discussion about finding a better way to filter out the jumps.

Did the suggested patch -- right now the problem seems to be gone. Two runs of timer-test didnt get any disphases.

[some_pinephone_user]

I did some new test. The RCU failre on deischarging battery logging battery levels. The results are pretty strange: https://dropmefiles.com/4IRLI
Sometimes the test-cycle completes without any error and sometimes with tens of thousand.

@smaeul , could you take a look into?

Your logs show time jumps that are smaller than those seen on any device before. They are not covered by the existing workaround in the kernel, so an additional kernel change is needed to resolve the errors you are seeing.

The quickest way to fix your time jumps is to apply this patch again, but instead of changing the constant from a 9 to an 8, change it from an 8 to a 7:

https://lore.kernel.org/linux-sunxi/2021...lland.org/

Decreasing the mask size could have a slight performance impact, so there has been some discussion about finding a better way to filter out the jumps.

Did the suggested patch -- right now the problem seems to be gone. Two runs of timer-test didnt get any disphases.

Is there some progress? Because the newer kernels without a patch still freeze randomly