There is no 4.9.1 yet. Are we talking about 4.9-1?

Any other errors?

4.9-1 indeed. I'm using my phone for this :)

It skips fsck, then the above error and then it drops me to an emergency shell.

Edit: before that, it says ata2: softreset failed (device not ready)

But it has done that since forever. I remember looking it up a year ago and back then it seemed harmless.

What version did you update from?

I don't recall, but I last did a full upgrade about 3 weeks ago so I guess 4.8.13, but it could be 4.8.12

Does lvm scan work at the emergency shell?

No such command. Would that mean the lvm hook somehow didn't get processed by mkinitcpio? I don't have a rescue disk at hand, but it'll be helpful to know the cause.

Exit from the emergency shell caused a kernel panic. However I noticed in the trace that raid1 and dm_raid modules are linked in. Is that new? I have previously added them to mkinitcpio.conf. Could there be a conflict there?

Excuse me, it should be lvm lvscan or lvm pvscan, but the lvm binary should definitely be present.

Nevermind that. Yes, that not only works but also gives the correct lv's. It warns about not using lvmetad, which I disabled a while ago. Stupid I didn't think of that, but from what I understand it only caches to avoid redundant scanning and is not really required.

Disabled how? The hook uses lvmetad.

Disabled in lvm.conf because it has a bug which causes it to hang during shutdown. I was not aware of the lvm hook using it; I thought it was just the systemd service (which is causing that bug). It has worked fine without it until now. If the hook requires it, it should give an error if it's disabled. I'll try to reenable it tomorrow when I have a rescue disk. Thanks for your help, I hope that'll fix it.

lvmetad being disabled was indeed causing the problem. I guess I'll have to look into that, because I seem to have misunderstood it's purpose. It's also fairly strange that lvmetad now only runs in initram, but because of the aforementioned bug, I still have the systemd service disabled. The latter is not causing any issues (except for a warning when doing lvm operations and them taking a bit more time).

But now I'm really confused, because the hook does not seem to read lvm.conf

It seems the issue was entirely different. When reenabling lvm in lvm.conf I also reinstalled lvm and the kernel to be sure. I think the lvm hook in initcpio had failed due to an issue with libreadline.so (see https://bbs.archlinux.org/viewtopic.php?id=219538). I probably missed it, because the initcpio itself completed without further error.