=== chris14_ is now known as chris14 [05:25] * alkisg rolls up sleeves, grabs ubuntu-23.04-beta-live-server-amd64.iso, and decides to spend the morning on this... wish me luck :) [05:37] alkisg, thanks, I'll also investigate a bit more on this today, let me know if you find anything.. and good luck! [05:45] Comparing jammy server with lunar server, with init=/bin/bash. [05:45] Original: jammy=91M, lunar=145M. So smaller diff than the desktop versions, but still significant. [05:45] With MODULES=dep, the initramfs size shrinked from 126M to 63M. Still lunar=145M, so it's not related to the initramfs size. [05:51] Purging the following packages shrinked the initrd to just 18M but made no difference to used RAM: linux-modules-extra-6.2.0-20-generic, amd64-microcode, intel-microcode === lifeless_ is now known as lifeless [06:26] Swapping the kernels (6.2 on 22.04 and 5.15 on 23.04) made no difference [06:40] alkisg, have you created a bug in launchpad for this issue already? [06:40] arighi: no, I don't know where to file it yet! [06:41] arighi, ok, just wanted to double check, I'll create one later [06:41] Currently I'm completely puzzled. I copied the whole /boot, and the difference is still there! I see an "unsupported ext4 FEATURE_C12", checking if it's related to the ext4 module and file system... [07:11] I added `copy_exec /bin/free /bin` in /usr/share/initramfs-tools/hooks, to copy the `free` command to the initramfs. [07:11] Now I'm testing with `break=top`, even before udev runs. There, jammy=35M, lunar=340M used RAM. [07:11] If I wasn't seeing this with my own eyes, I wouldn't believe the bug reporter... :) [08:07] arighi: I think I found the major difference, it's in what `free` reports. [08:07] With `break=bottom`, if I mount a jammy disk and `chroot /root free`, it says 65M, while with a lunar disk at that exact point, it says 500M. [08:07] So the "used" column is different, yet the "available" column is the same. [08:23] alkisg, hm.. interesting, how does /proc/meminfo look like? [08:23] this would tell us if the difference is actually in free showing a different result, because theoretically it should read all the mem info from /proc/meminfo [08:35] I have to do something urgent, back later on :/ [08:37] alkisg, ack, no problem, take your time [09:17] arighi: https://termbin.com/4mjk [09:27] alkisg, I see only one /proc/meminfo, that is in lunar I assume, it'd be nice to see /proc/meminfo also in the jammy case [09:29] arighi: that test was: "boot jammy with break=bottom; so only the initramfs is loaded; then chroot to jammy and run free, or chroot to lunar and run free" [09:29] So it's a different interpretation of the SAME /proc/meminfo file [09:29] alkisg, moreover I'm wondering if this is related to: https://gitlab.com/procps-ng/procps/-/commit/2184e90d2e7cdb582f9a5b706b47015e56707e4d [09:29] -ubottu:#ubuntu-kernel- Commit 2184e90 in procps-ng/procps "library: modify calculation of 'MEMINFO_MEM_USED' item" [09:29] alkisg, ahh ok got it! thanks for the clarification [09:29] That's probably it. So I think it's a false alarm. I did read the procps news and changelog, that commit didn't stand out, thanks for pointing me to it [09:30] yeah I'm pretty sure jammy's procps doesn't have that commit [09:31] Based on my tests so far, I think 22.10 doesn't have it either [09:31] alkisg, according to the git log the commit has been introduced in procps v4.0.1 [09:32] 22.10 is using a 3.3.17 [09:32] OK everything matches; that's it; thanks a lot for your help on this :) [09:32] and lunar 4.0.3-1ubuntu1 [09:32] so yeah, I think it all makes sense now :) [09:33] thank you for reporting this alkisg [09:33] 👍️ [10:53] arighi, its autopkgtests.ubuntu.com not having them [10:56] LocutusOfBorg, hm... it doesn't have EFI? [10:56] who knows? :) [10:57] so the kernel is right at complaining :) [11:12] meh, who can know this? [11:12] in Debian autopkgtests are EFI enabled I would say