Created attachment 292559 [details] photo of terminal after crashing the kernel I'm running Arch Linux on an Ryzen 2700x. kdump doesn't work with more recent kernels. I've bisected to the following commit: be62dbf554c5b50718a54a359372c148cd9975c7 iommu/amd: Convert AMD iommu driver to the dma-iommu api Booting the kernel with kexec right away (without the -p flag) still works on most recent kernel versions. The most recent version tried is v5.9-rc5. I use the following command to test. $ kexec /boot/vmlinuz-linux-kdump --initrd=/boot/initramfs-linux-kdump.img --reuse-cmdline Trying to use kdump doesn't work. I've set up the kexec crash kernel and use sysrq to crash the kernel. $ kexec -p /boot/vmlinuz-linux-kdump --initrd=/boot/initramfs-linux-kdump.img --append="root=UUID=1558db24-61ba-4fcd-ae27-63571a6243c2 single irqpoll nr_cpus=1 reset_devices" $ cat /sys/kernel/kexec_crash_loaded 1 $ echo c | sudo tee /proc/sysrq-trigger The attached kdump.jpg is a photo of the terminal after crashing the kernel. I don't have any other way to get the log after crashing the kernel.
Created attachment 292561 [details] v5.4 kdump config
Hello, I found my crash kernel also not working anymore lately. I am running a Debian testing and found also that the 5.4 kernel was the last working as expected. I found that in my case the crash kernel might really boot, but cannot find the root partition for some reason. Maybe you could boot with kernel parameter "modprobe.blacklist=amdgpu" and retest (if you have a amd gpu)? That way the screen stays with some efi framebuffer driver and the output of the crash kernel was shown on the screen. Further in my case I tried also booting the system with the crash kernel and the crash initrd but with regular kernel parameters. Because that worked I guess in my case all needed modules are really available - but cannot work for some other unknown reason. Could you confirm that too?
(In reply to Bernhard Übelacker from comment #2) > Hello, I found my crash kernel also not working anymore lately. > I am running a Debian testing and found also that the 5.4 kernel > was the last working as expected. > I found that in my case the crash kernel might really boot, but cannot > find the root partition for some reason. I mis-typed the title. It's including 5.4 that's broken for me. The bisected commit lies between 5.4-rc3 and 5.4. > > Maybe you could boot with kernel parameter "modprobe.blacklist=amdgpu" > and retest (if you have a amd gpu)? > That way the screen stays with some efi framebuffer driver > and the output of the crash kernel was shown on the screen. I'm using a radeon GPU. It is RV730 XT [Radeon HD 4670]. I tried to blacklist the radeon module. Nothing from the kdump kernel displays on screen. I suspect the kdump kernel doesn't boot far enough to display anything. > > Further in my case I tried also booting the system with the crash kernel > and the crash initrd but with regular kernel parameters. > Because that worked I guess in my case all needed modules are > really available - but cannot work for some other unknown reason. > Could you confirm that too? It doesn't work with regular kernel parameters which is just specifying the root file system and loglevel. $ cat /proc/cmdline BOOT_IMAGE=/boot/vmlinuz-linux-lts root=UUID=1558db24-61ba-4fcd-ae27-63571a6243c2 rw loglevel=7