kernel 5.2.1 and 5.1.16 regression bugs

Since about 5.1.7, the kernel worked well with amdgpu both in terms of suspend/resume and overall stability, until 5.1.15. In 5.1.16, suspend/resume problems resurfaced and continues with 5.2.1. 5.2.1 actually is worse with severe screen corruption, both with and without compositor enabled in KDE. The screen corruption happens with both radeonsi and amdgpu xorg drivers.

With 5.1.15, both radeonsi and amdgpu drivers (xorg) work well, good stability overall. I tried the 5.2.2 in kernel-stable repo but that wouldn’t boot because of secure boot ?!

In the first boot after installing the 5.2.2 kernel, there should have been a blue pop-up, asking if you want to enroll a key. If you had said “yes” and given the root password, that should have allowed the 5.2.2 kernel to pass secure-boot.

I see the blue screen after every (new) kernel install- it says to press any key to manage something. I take no action and it continues to the boot screen. It seems that the keys are automatically registered and I only have to take action to de-register them.

Isn’t that correct ? Are “unofficial” kernels packaged differently ?

I’m not sure of the full story.

There was a period where there were frequent blue screen popups for Tumbleweed. I think that had to do with testing a new feature.

There is a directory “/etc/uefi/certs”. As far as I know, the certs that are needed for secure-boot are kept there.

I’m not sure what happens with a kernel install, but I think it installs any needed cert in that directory. And, after install, a completion script runs that checks whether MokManager knows about all needed certs. And if not, then it runs “mokutil” to request that a certificate be imported, using “–root-pw”. So you have to give the root password to complete the import.

Kernels that come with Tumbleweed should already be signed by a known key. Kernels installed from the “Kernel:/stable” repo are signed with a different key and that key import will probably show up as a blue screen on the next boot.

I imported the OBS cert (different from opensuse cert) using mokutil. Then I had to add it in mok manager. So it boots 5.2.3 now. But it has same problem with the graphics corruption. Screen corrupted each time there’s an update by any application- not usable.

If you have not already done so, then I suggest that you add “oldest” to the “multiversion.kernels” line in “/etc/zypp/zypp.conf”. As in:

multiversion.kernels = oldest,latest,latest-1,running

That way, the oldest kernel won’t be automatically deleted. You can manually remove it when you have a newer kernel with everything working.

This appears to be an upstream bug in kernel 5.2 and also present in 5.3rc2 (at least).

Apparently no linux developer uses AMD graphics or at least raven ridge graphics or they would have encountered this issue and fixed it.

There’s no bug here with Vega10 GPU with secure boot enabled.

/etc/uefi/certs
4659838C.crt
EA1FD3EF.crt

Both Kernel Repos are enabled here and there is/was no issues to report. Kernel 5.1.21 has reached End Of Life (EOL) level.
https://lkml.org/lkml/2019/7/28/19

rpm -qa | grep -i kernel
kernel-firmware-20190712-272.1.noarch
kernel-default-5.3.rc3-1.1.g571863b.x86_64
kernel-default-5.2.6-1.1.g187170c.x86_64
mokutil  --sb-state
SecureBoot enabled

Our laptop booted in MoK only once after the fresh install (20190730), immediately after enabling Stable Kernel Repo. There is no password to deal with here, only clicking on continue to boot (already highlight) is enough to enroll the new key.

You boot in MoK Manager (Blue Screen) only and only if secure boot is enabled.

rpm -qa | grep -i amdgpu
libdrm_amdgpu1-2.4.99-1.1.x86_64
xf86-video-amdgpu-19.0.1-1.4.x86_64
rpm -qa | grep -i xorg
xorg-x11-libX11-ccache-7.6-21.5.x86_64
xorg-x11-fonts-7.6-37.1.noarch
xorg-x11-Xvnc-module-1.9.0-7.4.x86_64
xorg-x11-fonts-core-7.6-37.1.noarch
xorg-x11-server-wayland-1.20.5-1.2.x86_64
xorg-x11-server-1.20.5-1.2.x86_64
xorg-x11-Xvnc-1.9.0-7.4.x86_64
xorg-x11-server-extra-1.20.5-1.2.x86_64
xorg-x11-essentials-7.6_1-16.8.noarch
xorg-x11-driver-video-7.6_1-17.5.x86_64
inxi -G
Graphics:  Device-1: AMD Raven Ridge [Radeon Vega Series / Radeon Vega Mobile Series] driver: amdgpu v: kernel 
           Display: x11 server: X.org 1.20.5 driver: amdgpu FAILED: ati unloaded: fbdev,modesetting,vesa 
           resolution: <xdpyinfo missing> 
           OpenGL: renderer: AMD RAVEN (DRM 3.33.0 5.3.0-rc3-1.g571863b-default LLVM 8.0.1) v: 4.5 Mesa 19.1.3 

Regards,

I have another 2500U based notebook (mate book D) where I have ubuntu daily installed with kernel 5.2 running stock gnome desktop of ubuntu. Same problem- tearing and screen corruption. I also see messages that some raven related firmware files are missing (on ubuntu).

I was about to open bug reports when I saw some related bugs-

https://bugzilla.kernel.org/show_bug.cgi?id=204181
https://bugzilla.kernel.org/show_bug.cgi?id=204241
https://bugzilla.kernel.org/show_bug.cgi?id=203471

The suspend/resume failure, present till 5.0, fixed in 5.1 has returned in 5.2

@mvaar

Sleep and Hibernate won’t resume, stays black and magic key has no effect (Kernel5.3). Sleep works in Win Ten, but in both cases we are not using them. Lock and logout are working properly. Latest snapshot (August 5th) has Kernel 5.2.5 in it with KDE 5.16.4.

We don’t have the screen issue that you describe. If we enlarge Swap to RAM size, sometimes it works. Bootline is tweaked with iommu=soft if this can help. Are you using a Kernel parameter(s) at boot?

I am using “workqueue.power_efficient” parameter as by default most distros use that setting. My swap partition is already larger than memory and in fact, with 5.1.7, even hibernate worked.

I tried kernel boot param iommu=soft and that fixed it ! No screen corruption and suspend/resume works with 5.2.7. I didn’t try hibernate yet.

Thanks.