Problem with screen flickering - i915

Rydberg95 · 2019-01-10 15:55:36

Hello forum,

I've been off and on Arch Linux for the last couple of my yeas, and therefore consider myself a beginner.

Now, yesterday I bought a new laptop and decided to commit to running Arch Linux as my daily driver. The relevant spec is as below. Please tell me if more of the hardware is relevant, and I'll be happy to provide it.

Specifications
Clevo N750WU
Intel Core i7-8550U
8 GB Kingston DDR4 SODIMM 2400 MHz
Samsung 970 EVO 500 GB
Intel AC-9260 WiFi module 1730Mbps
Running the latest version of Arch Linux

The problem:
After boot (without kernel parameters) the screen becomes black and “flickers”, showing the login prompt a few times a minute, for around one tenth of a second.

The only error I get is that sometimes at boot, the following message is displayed:
drm:intel_cpu_fifo_underrun_irq_handler [i915]] *ERROR* CPU pipe A FIFO underrun

Workarounds:
1. After some research, I understood that there was something wrong with the i915 module, and when booting with kernel parameter i915.modeset=0, there is no flickering. However, this makes me unable to use Xorg, which I want to be able to.
2. Booting without the kernel parameter above, and having the NetworkManager.Service disabled. As soon as I start it, the flickering behavior returns. The same goes when using wpa_supplicant/dhcpd, and running the dhcpcd command afterwards starts the flickering.

At first I thought it was the display drive alone that was causing the problem, but number two above leads me to believe it’s something else. Worth mentioning is that I tried running Ubuntu, and it does not have the same problem, even when connected to a wifi.

I realize that I will probably need to post some logs, however, I’m unsure as of which ones.

Related links:
https://wiki.archlinux.org/index.php/in … flickering
Above mentions power saving, and booting with the i915.enable_psr=0 flag does not make any difference.

It would be greatly appreciated to get any help with fixing this issue.

Trilby · 2019-01-10 16:22:53

There was another thread with identical sounding symptoms also with a very odd/unintuitive trigger/workaround:
https://bbs.archlinux.org/viewtopic.php?id=243303

I'd speculate that there may be an issue with i915 that is revealed under a range of other seemingly unrelated conditions (network connection, RAID array, etc).

Rydberg95 · 2019-01-10 17:09:29

Trilby wrote:

There was another thread with identical sounding symptoms also with a very odd/unintuitive trigger/workaround:
https://bbs.archlinux.org/viewtopic.php?id=243303
I'd speculate that there may be an issue with i915 that is revealed under a range of other seemingly unrelated conditions (network connection, RAID array, etc).

Thank you for your reply.

Do you know if there is anything to do about it? Maybe raising it to be patched?

I don't know really how this works, but maybe I'm able to use the same module that Ubuntu uses, since it seems to work?

V1del · 2019-01-10 17:14:44

If you use xf86-video-intel remove it, if it is still reproducible try the LTS kernel, maybe 4.20 has something fishy going on. Using the module from Ubuntu isn't really feasably doable (you could use the entire kernel, but there are saner avenues to go down in Arch land before we get that desparate)

Last edited by V1del (2019-01-10 17:15:43)

Rydberg95 · 2019-01-10 20:05:08

V1del wrote:

If you use xf86-video-intel remove it, if it is still reproducible try the LTS kernel, maybe 4.20 has something fishy going on. Using the module from Ubuntu isn't really feasably doable (you could use the entire kernel, but there are saner avenues to go down in Arch land before we get that desparate)

Thank you for your reply.

Do you mean installing the xf86 driver and removing i915, or removing xf86 and using i915?

seth · 2019-01-10 22:00:56

You're not going to see much w/o i915 (the kernel module), the only thing you can toggle is the X11 driver (intel ./. modesetting)

pacman -Qs xf86-video

blochl · 2019-01-11 05:59:29

It sounds like this: https://bugs.freedesktop.org/show_bug.cgi?id=103229

Try adding `intel_idle.max_cstate=4` to the kernel parameters.

Rydberg95 · 2019-01-11 07:45:18

seth wrote:

You're not going to see much w/o i915 (the kernel module), the only thing you can toggle is the X11 driver (intel ./. modesetting)
pacman -Qs xf86-video

I see. Thank you for explaining. However, it didn't make any difference with the screen flickering.

blochl wrote:

t sounds like this: https://bugs.freedesktop.org/show_bug.cgi?id=103229
Try adding `intel_idle.max_cstate=4` to the kernel parameters.

This seems to work! I'm able to run NetworkManager at boot and can start X without any screen flickering. Thank you very much!

Now, regarding this kernel parameter, I understand that it prevents the CPU from going in to lower power states. This is, of course, not optimal for using a laptop on battery power, and seems like a workaround.

This solves my short term problem and makes me able to user my laptop (which I am very happy for), however, is there any way that this issue can be fixed within the driver/kernel(?), or is that too much to hope for?

Rydberg95 · 2019-01-11 09:00:49

An update:

After some reading up on the intel_idle.max_cstate kernel parameter, I get the understanding that the if I set a lower value, the CPU is given the possibility to do more in terms of power saving (correct me if I'm wrong).

So, I tried decreasing the value and I there seems to be no difference between setting intel_idle.max_cstate=4 and intel_idle.max_cstate=1. However, setting the value 0 makes the flickering return, which is reasonable since the default value is 0. Therefore, I will boot with the kernel parameter intel_idle.max_cstate=1 from now on.

V1del · 2019-01-11 09:18:27

You're wrong It's the other way around, the higher the value, the deeper the sleep state. 0 just happens to be a special case, which disables intel_idle and uses acpi_idle instead. So if you can't reproduce with higher max_cstate values that would be better for power consumption. Possible ranges are 1 to 9

Last edited by V1del (2019-01-11 09:20:21)

progandy · 2019-01-11 09:24:03

With c-states a higher value does more power saving, so use the highest value that is stable. You have probably confused c-states and p-states.

https://software.intel.com/en-us/articl … e-c-states

Last edited by progandy (2019-01-11 09:25:57)

Rydberg95 · 2019-01-11 13:57:36

V1del wrote:

You're wrong It's the other way around, the higher the value, the deeper the sleep state. 0 just happens to be a special case, which disables intel_idle and uses acpi_idle instead. So if you can't reproduce with higher max_cstate values that would be better for power consumption. Possible ranges are 1 to 9

Of course, why would it be so simple...

It seems like 4 is the highest it can go before flickering starts, so I guess that's what I'll use.

Arch Linux

#1 2019-01-10 15:55:36

Problem with screen flickering - i915

#2 2019-01-10 16:22:53

Re: Problem with screen flickering - i915

#3 2019-01-10 17:09:29

Re: Problem with screen flickering - i915

#4 2019-01-10 17:14:44

Re: Problem with screen flickering - i915

#5 2019-01-10 20:05:08

Re: Problem with screen flickering - i915

#6 2019-01-10 22:00:56

Re: Problem with screen flickering - i915

#7 2019-01-11 05:59:29

Re: Problem with screen flickering - i915

#8 2019-01-11 07:45:18

Re: Problem with screen flickering - i915

#9 2019-01-11 09:00:49

Re: Problem with screen flickering - i915

#10 2019-01-11 09:18:27

Re: Problem with screen flickering - i915

#11 2019-01-11 09:24:03

Re: Problem with screen flickering - i915

#12 2019-01-11 13:57:36

Re: Problem with screen flickering - i915

Board footer