Why are 6GB unavailable on 32GB? (Debian 12)

(This is updated with each new information) My HP Proliant ML350 Gen 10 server has 32 GB RAM, that seem correctly detected. But only 26GB are available, according to free/htop/proc.meminfo…

free -m
               total        used        free      shared  buff/cache   available 
Mem:           31755        6185       25533           6         436       25570 
Swap:            976           0

MemTotal:       32518056 kB
MemFree:        26173304 kB
MemAvailable:   26250792 kB

OS : Debian 12 bookworm Other details and command outputs are too long for this forum, I've put them here: https://semestriel.framapad.org/p/phqm1kqhph-afqg?lang=fr Here are the different tests/ideas/options: * These are not buffers. * tmpfs system mounts are almost empty. * I have no huge pages. * I have killed or desinstalled more or less anything consuming memory that I could find. * I have no graphic card (VGA), and only one CPU. * No ZRAM * No ZFS, only ext4 * I've removed the NVMe disc, the SSD drives. The OS lies on a 128 GB SDCard. * A Live Debian with nothing installed shows the same already-used 6GB (with all graphical interfaces off) * Same problem in emergency mode. * An old Kali on LiveUSB avec a 5.10 kernel does not show this. But a new Kali with Linux 6.12 shows the same problem as Debian 12. Kali is Linux-based, so this is coherent. This confirms that this the RAM is not eaten by a software that I have installed. * I've erased the LVM partitions, removed the disks (physically) * I've deactivated ILO (the remote administration interface from HP): /proc/iomem does not show any memory relase. But it should not amount to more thant 300MB. I have discovered things like /proc/iomem or /proc/modules far from my area of expertise, without finding huge amounts of hidden memory. I've found [this thread](https://unix.stackexchange.com/a/730970) on memory used by the kernel, but this is a 1.5% loss, not 20%! I found [this other thread on "non-cached kernel dynamic memory"](https://unix.stackexchange.com/questions/62066/what-is-kernel-dynamic-memory-as-reported-by-smem) that does not explain anything in fact. * By comparison, an (empty) Vagrant machine with 32GB RAM shows 32'476'804 available. * And my laptop with 64 GB RAM, Debian 12 too, can go down to less than 2GB used in emergency mode. I really would like to know where these 6GB are gone, and if I can recover a part of it. And yes, the system really feels like it has only 25-26GB. As a test, I've created 24GB of static huge pages (more seems impossible), the free/available memory becomes almost nothing. (I've found this problem in a test, where a PG cluster with a 24GB shared_buffers collapses, with massive IOwait on the slow root partition.) Thank you for any advice.