Extreme stuttering with i915.enable_psr=1

@qdrop, Can you please add "drm.debug=0x1e log_buf_len=2M" into your kernel parameters and provide dmesg here? Also would it be possible to try with latest drm-tip?

ok, I added to dmesg output here:

journalctl-dmesg.log

I have to check on how to build the kernel properly first; maybe the dmesg already helps you enough?

Thank you for the log. Yes, dmesg already helps me in understanding your problem. Can you please try setting i915.psr_safest_params=1 kernel parameter together with i915.enable_psr=1 to see if that helps on your issue?

Alright; I have set the parameters accordingly; the issue is fixed too this way. But I experience some flickering on the internal screen now. All couple of minutes. It's not very bad but I can notice it.

How can I continue supporting you in narrowing down the issue?

@qdrop thank you for your testing. I updated my bisection set here: https://patchwork.freedesktop.org/series/117996/. If you could test patches 2. and 3. separately?

I have also uploaded fix candidate into trybot. Addition to my previous request if you have time you could also test that one: https://patchwork.freedesktop.org/series/118018/ Thank You in advance.

Alright, I tested your patches:

[qdrop@qdrop-fedora-pro9i drm-tip]$ git apply --stat v2-2-3-drm-i915-psr-Use-maximum-fast-wake-lines.patch
 drivers/gpu/drm/i915/display/intel_psr.c |    2 ++
 1 file changed, 2 insertions(+)

-> This did work without i915.enable_psr=0 as kernel parameter. Stuttering gone.

[qdrop@qdrop-fedora-pro9i drm-tip]$ git apply --check v2-3-3-drm-i915-psr-Use-maximum-io-wake-lines.patch
error: patch failed: drivers/gpu/drm/i915/display/intel_psr.c:873
error: drivers/gpu/drm/i915/display/intel_psr.c: patch does not apply

-> I wasn't even able to apply the patch.

[qdrop@qdrop-fedora-pro9i drm-tip]$ git apply --stat drm-i915-psr-Adjust-time-values-used-for-io-and-fast-wake.patch
 drivers/gpu/drm/i915/display/intel_psr.c |   10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

-> This did not work without i915.enable_psr=0: The issue with the stuttering remained

Please note that I did not build the Nvidia-Driver (for these kernels) - it did fall back to nouveau. But as I'm only using the Intel iGPU for rendering my DE I don't think this has anything to do with our issue.

@qdrop , Thank you very much for your help here. Were you testing using drm-tip or something else?

Yes, I followed this howto here: https://www.intel.com/content/www/us/en/docs/graphics-for-linux/developer-reference/1-0/build-guide.html

git clone git://anongit.freedesktop.org/drm-tip

My builds resulted in a 6.4.X kernel.

@qdrop , ok. Any chance you could still give a try to following patch:

https://patchwork.freedesktop.org/series/118412/

?

Sure, I just tested it.

It seems to work; but I tried (by chance) to activate the mirroring mode for my two screens (internal & external) and the issue returned on the internal screen and after a couple of seconds the system froze entirely. I was not even able to switch to a TTY shell.

@qdrop, that is a interesting finding. Can you check again how i915.enable_psr=0, i915.psr_safest_parameters=1 and

[qdrop@qdrop-fedora-pro9i drm-tip]$ git apply --stat v2-2-3-drm-i915-psr-Use-maximum-fast-wake-lines.patch
 drivers/gpu/drm/i915/display/intel_psr.c |    2 ++
 1 file changed, 2 insertions(+)

each of them separately are behaving with this setup?

This has always been the case. The external monitor never had this issue; only the internal one with 120hz set as refresh rate.

The issue arised when enabling mirroring as I wasn't able to set the refresh per screen manually.

I'm away for a week now, but I'm ready to do further testing when returning. Maybe we can set up some test matrix?

@qdrop , I have now merged https://patchwork.freedesktop.org/series/118517/ into drm-tip. Can you please try to reproduce this with latest drm-tip which contains https://patchwork.freedesktop.org/series/118517/?

Hi. I just did this and rebooted without i915.enable_psr=0 as kernel parameter. Unfortunately the issue persists.

ok, thank you for testing this. Did you had external monitor connected? Are you still able to reproduce it if you unplug it?

Yes, it does not matter whether the external screen is attached or not: The stuttering persists on the internal screen. The external screen works properly.

Thank you. Can you please ensure that you have following patch in the drm-tip you are using?:

https://patchwork.freedesktop.org/series/118517/

is i915.psr_safest_params=1 still fixing the issue when you are using drm-tip?

This patch is not applicable:

[qdrop@qdrop-fedora-pro9i drm-tip]$ git apply --check v3-drm-i915-Use-18-fast-wake-AUX-sync-len.patch
error: patch failed: drivers/gpu/drm/i915/display/intel_dp_aux.c:129
error: drivers/gpu/drm/i915/display/intel_dp_aux.c: patch does not apply

This line of code is already in the state the patch would apply. In this state I experience the stuttering without i915.enable_psr=0.

Setting i915.psr_safest_params=1 resolves the issue though.

qdrop@qdrop-fedora-pro9i ~]$ uname -r
6.4.0-rc5-drmtip+

Ok, it seems the fast wake aux sync adjustment didn't fix your issue after all. Thank you for testing these again.

"Whenever I try to activate 120hz on the internal screen, there is extreme stuttering on it": Is there any difference if you try to use 120Hz from the boot already? I.e. set it to 120Hz and then reboot the machine.

No, I've been using the 120hz setting for all of my recent tests. Please note, however, that when I reboot my computer, GDM3 (login / display manager) renders at 60hz. The screen settings are only applied after logging in / starting the Wayland session.

You can try copying .config/monitors.xml from the home directory of your user to that of the gdm user, e.g. /var/lib/gdm/ or /var/lib/gdm3/.

I did this, but I doubt GDM runs with 120hz now. Unchanged behaviour: Only i915.psr_safest_params=1 resolves the issue.

@qdrop, thank you again for you support here. Can you please provide me PSR2_CTL and DDI_AUX_CTL register values from the setup where you can reproduce the issue (i.e. drm-tip with i915.enable_psr=1 and without i915_psr_safest_params=1)? You can obtain the values by install intel-gpu-tools and then use intel_reg tool from command line:

sudo intel_reg read 0x64010 0x60900

wow, I think I assumed something wrong (I know, assuming is dangerous ;-)): i915.enable_psr=1 and no parameter equals the same / the default value. But apparently I was wrong:

Setting i915.enable_psr=1 (without i915_psr_safest_params=1) leads to a working system.
Setting i915.enable_psr=0 (without i915_psr_safest_params=1) leads to a working system.
Not setting i915.enable_psr at all (without i915_psr_safest_params=1) leads to the issues.
Not setting i915.enable_psr at all and setting i915_psr_safest_params=1 leads to a working system.

I'm currently running 6.3.5-200.fc38.x86_64. Doing this in the default config (without setting i915.enable_psr at all) works properly.

[qdrop@qdrop-fedora-pro9i ~]$ sudo intel_reg read 0x64010 0x60900
Warning: register spec not found in '/usr/share/igt-gpu-tools/registers'. Using builtin register spec.
                                    (0x00064010): 0x2c2002ff
                                    (0x00060900): 0x80004a26

@qdrop, thank you for your response. Based on AUX_CTL value you are not using latest drm-tip. Seems to be still using old value for fast wake sync pulse count (24 pulses). We should expect seeing "(0x00064010): 0xXXXXX23f"

Please try to boot latest drm-tip with i915.enable_psr=1 and without i915_psr_safest_params=1. Ensure you are seeing the issue and then take the register snapshots using following spell:

uname -r && cat /proc/cmdline && sudo intel_reg read 0x64010 0x60900

output will ensure you have booted the right kernel and your kernel parameters are as expected. Thank you in advance.

Setting i915.enable_psr=1 (without i915_psr_safest_params=1) leads to a working system. Setting i915.enable_psr=0 (without i915_psr_safest_params=1) leads to a working system. Not setting i915.enable_psr at all (without i915_psr_safest_params=1) leads to the issues. Not setting i915.enable_psr at all and setting i915_psr_safest_params=1 leads to a working system.

These are somehow conflicting. Maybe you could recheck and on every boot ensure you environment is as assumed by checking used kernel and it's cmdline:

uname -r && cat /proc/cmdline

If Fedora have disabled PSR by default in their kernel by setting enable_psr default as 0 might do some tricks to us here...

There you go:

[qdrop@qdrop-fedora-pro9i ~]$ uname -r && cat /proc/cmdline && sudo intel_reg read 0x64010 0x60900
6.4.0-rc6+
BOOT_IMAGE=(hd0,gpt5)/vmlinuz-6.4.0-rc6+ root=UUID=770747da-b910-4660-9326-edb9806b0379 ro rootflags=subvol=root rd.driver.blacklist=nouveau modprobe.blacklist=nouveau rd.luks.uuid=luks-f0e7cab1-0490-4186-89fb-bf904d295aa8 rhgb quiet rd.driver.blacklist=nouveau modprobe.blacklist=nouveau i915.enable_psr=1
Warning: register spec not found in '/usr/share/igt-gpu-tools/registers'. Using builtin register spec.
                                    (0x00064010): 0x2c10023f
                                    (0x00060900): 0x00004811

This configuration is working as expected. Removing i915.enable_psr=1 leads to the issues.

@qdrop, thank you. This seems to have now "(0x00064010): 0xXXXXX23f". Can you please still provide me output of "uname -r && cat /proc/cmdline" with same setup except leave that i915.enable_psr=1 out. I.e. the problematic setup.

Here is the same command in a problematic setup (lagging internal screen):

[qdrop@qdrop-fedora-pro9i ~]$ uname -r && cat /proc/cmdline && sudo intel_reg read 0x64010 0x60900
6.4.0-rc6+
BOOT_IMAGE=(hd0,gpt5)/vmlinuz-6.4.0-rc6+ root=UUID=770747da-b910-4660-9326-edb9806b0379 ro rootflags=subvol=root rd.driver.blacklist=nouveau modprobe.blacklist=nouveau rd.luks.uuid=luks-f0e7cab1-0490-4186-89fb-bf904d295aa8 rhgb quiet rd.driver.blacklist=nouveau modprobe.blacklist=nouveau
Warning: register spec not found in '/usr/share/igt-gpu-tools/registers'. Using builtin register spec.
                                    (0x00064010): 0x2c20023f
                                    (0x00060900): 0x80004a26

Let me know if you have doubt about my kernel configuration. I'm pretty sure I managed to build and boot the drm-tip kernel properly.

To enable / disable flags I simply edit the boot command line in Grub before booting. cat /proc/cmdline reflects that this is working afaik.

@qdrop , thank you again. Now it seems to be as expected. I.e. 0x64010 being 0xXXXXX23f. With this setup could you please take dmesg and attach it here. Please ensure you have again "drm.debug=0x1e log_buf_len=2M" in your kernel command line.

[qdrop@qdrop-fedora-pro9i ~]$ uname -r && cat /proc/cmdline && sudo intel_reg read 0x64010 0x60900 && sudo journalctl -o short-precise -k > dmesg.log
6.4.0-rc6+
BOOT_IMAGE=(hd0,gpt5)/vmlinuz-6.4.0-rc6+ root=UUID=770747da-b910-4660-9326-edb9806b0379 ro rootflags=subvol=root rd.driver.blacklist=nouveau modprobe.blacklist=nouveau rd.luks.uuid=luks-f0e7cab1-0490-4186-89fb-bf904d295aa8 rhgb quiet rd.driver.blacklist=nouveau modprobe.blacklist=nouveau drm.debug=0x1e log_buf_len=2M
Warning: register spec not found in '/usr/share/igt-gpu-tools/registers'. Using builtin register spec.
                                    (0x00064010): 0x2c20023f
                                    (0x00060900): 0x80004a26

dmesg.log

This thing got big quite quickly ;-)

Thank you @qdrop. I went through the log you provided. To obtain some more information on what is happening I created following debug patch : https://patchwork.freedesktop.org/series/119364/

Can you please apply it and take again dmesg? Please use same setup as with previous dmesg snapshot except that patch being applied.

Alright, I downloaded the .patch-file, applied it to the same drm-tip sources, rebuild the kernel and modules, installed them, added the same arguments to the cmdline, rebooted and piped dmesg into this file:

dmesg.log

Hope that helps.

@qdrop , Unfortunately debug traces I added are not visible in the attached dmesg. I'm expecting seeing "Calculated io wake lines = X", "Calculated fast wake lines = X" and "EDP_PSR2_CTL = 0xXXXXX" traces in the logs. I see build stamp is today, maybe something else went wrong?

oh, excuse me, I patched the sources wrongly. The debugging lines are now existing in the dmesg:

dmesg.log

@qdrop , thank you very much! Can you please apply https://patchwork.freedesktop.org/series/119436/ on top of previous patch. Then test if you can still see stuttering. Also please capture dmesg again and attach here. Keep kernel command line as it was with previous dmesg capture.

voila, the issue seems resolved with this patch - is this possible?

[qdrop@qdrop-fedora-pro9i ~]$ uname -r && cat /proc/cmdline && sudo intel_reg read 0x64010 0x60900 && sudo journalctl -o short-precise -k > dmesg.log
6.4.0-rc6+
BOOT_IMAGE=(hd0,gpt5)/vmlinuz-6.4.0-rc6+ root=UUID=770747da-b910-4660-9326-edb9806b0379 ro rootflags=subvol=root rd.driver.blacklist=nouveau modprobe.blacklist=nouveau rd.luks.uuid=luks-f0e7cab1-0490-4186-89fb-bf904d295aa8 rhgb quiet rd.driver.blacklist=nouveau modprobe.blacklist=nouveau drm.debug=0x1e log_buf_len=2M
Warning: register spec not found in '/usr/share/igt-gpu-tools/registers'. Using builtin register spec.
                                    (0x00064010): 0x2c20023f
                                    (0x00060900): 0x9000da26

dmesg.log

@qdrop, yes it is. Thank you very much for your help here. I will send the patch out. Do you mind adding your Tested-by tag there? What email address shall I use?

sure, you can use tg-linux-contributions@q-drop.com. Thank you for your support; feel free to reach out for other tests ;-).

mentioned in commit Mesa_CI/repos/linux@2d6f2f79

mentioned in commit nouveau@b29a20f7

closed with commit b29a20f7

mentioned in commit nouveau@c16e79e2

reopened

mentioned in commit lyudess/linux@ef0af9db

mentioned in commit nouveau@5311892a

This is now fixed. Thank you for the report and testing support here.

closed

mentioned in issue #10168

Extreme stuttering with i915.enable_psr=1

Affected version

Bug summary

Child items ...

Activity

Admin message

Extreme stuttering with i915.enable_psr=1

Affected version

Bug summary

Activity