=== chris14_ is now known as chris14 [09:45] Boot 16.04 64bit with init=/bin/bash, then `free` => 49M used. For 22.04 => 86M. For 23.04 => 300M. [09:45] How can I find out what in the kernel configuration takes 200+ more MB RAM in 23.04? I don't see anything weird in ps or in lsmod... [09:51] alkisg, `slabtop -s c` maybe can tell if the kernel itself is using too much memory in some slab caches, moreover `free -m` would help to understand if most of the memory is page cache or actually allocated memory (I guess you already measured these values after dropping page cache, e.g., `echo 3 > /proc/sys/vm/drop_caches`?) [09:54] Output of `uname -a; free; lsmod; ps faux` for 22.10, i.e. 89M used: https://termbin.com/r5q8 [09:54] Output for 23.04, i.e. 344M used: https://termbin.com/ox2b [09:54] I'm looking at the "used" column, so caches are excluded; will check slabtop now [10:03] Output of slabtop -s c -o for 23.04: https://termbin.com/uwj8v [10:04] (dropping the caches makes no difference, it's too early in the boot process, no significant cache usage yet) [10:05] Output of slabtop -s c -o for 22.10: https://termbin.com/5v3s [14:28] Hello, I tried to run dkms autopkgtests on official autopkgtest machines [14:28] +EFI variables are not supported on this system [14:28] +/sys/firmware/efi/efivars not found, aborting. [14:29] I don't think "modprobe efivars" is sufficient to fix this, right? [14:29] see https://autopkgtest.ubuntu.com/results/autopkgtest-lunar/lunar/amd64/d/dkms/20230403_183258_890fd@/log.gz [14:30] on armhf I get a "better error" "Secure Boot not enabled on this system." [14:30] ^^ this is on arm64, while on armhf I get "Binary update-secureboot-policy not found, modules won't be signed" === cpaelzer_ is now known as cpaelzer [14:31] (same armhf on ppc64el and s390x) [14:31] and I don't think adding shim-signed will fix, right? [14:46] LocutusOfBorg, from a kernel perspective /sys/firmware/evi/efivars should be populated when the efivars fs is mounted, like `mount -t efivarfs none /sys/firmware/efi/efivars`, not sure why it doesn't show up on this test system... [14:47] I'm wondering what would happen if you try to mount that manually (not sure if it's possible) [15:12] arighi, thanks [15:12] the test has "needs-root" [15:12] soooooo maaaaaaaaaybe :D [15:13] testing on https://launchpad.net/~costamagnagianfranco/+archive/ubuntu/locutusofborg-ppa/+packages?field.name_filter=&field.status_filter=published&field.series_filter=lunar [15:19] well, at least on arm64 it says "Secure Boot not enabled on this system" [15:19] https://autopkgtest.ubuntu.com/results/autopkgtest-lunar-costamagnagianfranco-locutusofborg-ppa/lunar/arm64/d/dkms/20230411_151723_f2fca@/log.gz [16:21] mount: /sys/firmware/efi/efivars: mount point does not exist. [16:21] damn [17:34] Anything else I could try to see why the kernel needs 300+ MB RAM in 23.04, compared to 90- MB in 22.10? (see above for details and pastebins) [18:25] alkisg: it's a wild-guess shot in the dark, but check dmesg | grep -i crash on both? I have a very vague memory of the crashdump default being raised, or discussed to be raised?, or something similar anyway :) [19:46] sarnold: thanks; unfortunately there's nothing in `dmesg | grep -i crash` in 23.04 [19:53] (nor in 22.10) [20:12] alkisg: dang. well, not all my wild ass guesses work out :( [20:12] alkisg: how about /proc/meminfo on both? [20:12] Meh, it looks like it's an initramfs issue, not a kernel issue. I installed the 23.04 kernel in 22.04 and it doesn't have increased memory usage [20:13] I wonder if it's not freeing up the initrd memory. Looking more into it... [20:47] Heh. I put `break=bottom` and ran `mount --bind /proc /root/proc; chroot /root free`. 22.10 says 75M used, 23.04 says 730M used. [20:47] I think many 1GB cloud servers/VMs won't be able to boot at all anymore... [21:15] LocutusOfBorg, hm.. don't you have /sys/firmware/efi and /sys/firmware/efivars? [21:16] alkisg, I'm wondering if the initramfs is bigger because mkinitramfs started to put linux-modules-extra modules to the initramfs in 23.04...