Use Kdump to handle kernel panics and dumping core to the hard drive to troubleshoot them later? (split from Change Proposal "Enable Drm Panic")

Has Fedora considered the use of Kdump to handle kernel panics and dumping core to the hard drive to troubleshoot them later?

I understand it doesn’t always work, which DRM Panic could be a great backup. However, Kdump can automatically provide a crash dump, which could be displayed in the ABRT app, which can also upload these dumps to the Red Hat Bugzilla.

in fact, Red Hat Enterprise Linux (RHEL) already uses Kdump by default to facilitate these measures.

I’m just curious as to what the main opinion on it is, and what the experience has been overall with the community.

kdump and drm_panic are not mutually exclusive.

One thing that would be great, is if drm_panic can display a progress bar while kdump is dumping to the hard drive, but I’m not sure it’s technically feasible. (As kdump is another linux kernel, it can be risky to call functions from the “crashed” kernel to display the progress).

I didn’t test yet drm_panic with kdump, but I think the kdump kernel may overwrite the panic screen. This may depends on the graphic card.

Also kdump has some drawbacks, like needing to reserve some big chunk of memory, and that’s probably why it’s not enabled by default.

2 Likes

I split this into a separated ask-topic as it is not immediately related to the Change Proposal. @theoceanbreeze feel free to change the topic title or adjust your post, or add further posts of course :slight_smile:

2 Likes

One thing that would be great, is if drm_panic can display a progress bar while kdump is dumping to the hard drive, but I’m not sure it’s technically feasible. (As kdump is another linux kernel, it can be risky to call functions from the “crashed” kernel to display the progress).

I’ve heard someone on the internet was able to get the Plymouth service running on the crash kernel, although it looks like this isn’t default, at least when I tried it in a virtual machine.

There is also systemd-bsod that is currently being used only for boot failures. It may be possible to run the service from the crash kernel, although I have not seen too much research into this.

I didn’t test yet drm_panic with kdump, but I think the kdump kernel may overwrite the panic screen. This may depends on the graphic card.

I believe it is a good idea to look into this more, as I personally have not tested this yet. What I do know is that SystemD will output it’s progress to save the coredump onto the display.

Also kdump has some drawbacks, like needing to reserve some big chunk of memory, and that’s probably why it’s not enabled by default.

Is this a certain percentage of the current RAM in the system, or is it only reliant on how much RAM the kernel would need? It could be enabled if the user has enough memory, but more research should be done regarding current system requirements and their relevance to this drawback.

Thanks for the feedback and I look forward to hearing more of your insights. :slight_smile: