=== tafa2- is now known as tafa2 === zzz is now known as xxx === xxx is now known as xxxzzz === xxxzzz is now known as xxxyzzz [13:01] Hi guys. Should ufw dhcp-pd work out of the box? [13:06] https://pastebin.mozilla.org/9078945 [13:06] Is there anything wrong with this syntax. I am trying to allow one ip and block all but its blocking all ips (including the one I whitelisted) https://pastebin.mozilla.org/9078945 [14:43] coreycb: ping neutron-vpnaas could use a rebuild in the cloud-archive https://pastebin.com/hL9b25q7 [15:06] zul: neutron-vpn-agent has gone [15:10] as in the vpn bits run as an extension to the neutron-l3-agent in queens [15:27] argh...ok [21:41] I have a bunch of QEMU VMs with small RAM allocations (256M or less) but the VM's kernel seems to be missing ~20M leading to only ~236M visible/usable. Any way to find where the rest is gone? [21:42] "journalctl -kocat | grep Memory" gives me: Memory: 227764K/261624K available (8530K kernel code, 1309K rwdata, 3992K rodata, 1508K init, 1316K bss, 33860K reserved, 0K cma-reserved) [21:43] 261624K is pretty close to what the VM is given 262144K [21:43] sdeziel: says 33M reserved? [21:44] nacc: yeah, any way to influence this? Or learn what it's for? [21:44] is that a safety stash to be used only when under memory pressure? [21:44] sdeziel: can you `cat /proc/meminfo` to a pastebin? [21:45] https://paste.ubuntu.com/p/7mPvT9hf96/ [21:45] lol which give sa different number :) [21:45] my 236M visible from "free -m" is exactly the MemTotal [21:46] sdeziel: you've got about 20 M of reclaimable slab memory [21:46] I know that long ago there was a thing like reserved memory for DMA/ISA [21:46] which is relatively high for a small footprint system [21:46] SReclaimable > [21:47] sdeziel: yeah [21:47] i believe you can force that to reclaim by drop_caches [21:47] * sdeziel wonders how to reclaim [21:48] sdeziel: try echo 3 > /proc/sys/vm/drop_caches [21:48] brought it down to ~8M [21:48] the that is poking at something that's not MemTotal AFAICT [21:48] sdeziel: err, duh, you're right [21:49] sdeziel: can you pastebin the full dmesg? [21:49] I experimented with an even smaller VM (192M assigned) and was also missing 20M [21:50] https://paste.ubuntu.com/p/8ysM3kGX4t/ [21:52] sdeziel: i used to know this really well, but i'm a few years rusty now :/ [21:52] sdeziel: trying to remember [21:54] nacc: hehe, nothing urgent, I'm only trying to reclaim as much as I can to keep my super old VM server from running out of RAM [21:54] sdeziel: iirc, that memory is kernel pages [21:55] sdeziel: and i think that means you used something like 20M of bootmem or so [21:55] whcih can't be freed normally [21:55] I wonder what that means: e820: update [mem 0x00000000-0x00000fff] usable ==> reserved [21:55] hmm, bootmem you say [21:56] sdeziel: that's my recollection, the kernel's bootsrap memory [21:56] sdeziel: that's referring to the e820 map in firmware [21:56] your bios tells the kerenl physically what addresses are allowed, etc [21:57] OK [21:57] sdeziel: the kernel protects the first 64KB of memory from use, and actively scans it for corruption (to detect writes to the NULL page [21:57] TJ-: vm.mmap_min_addr right ? [21:58] TJ-: I tried shrinking it from 65536 -> 2048 and didn't get anything back [21:58] I also booted the linux-image-kvm flavor and got much less RAM taken away from me [21:59] sdeziel: I don't think so, no, the e820 0-0xFFF is reserving the first 4096 byte page for safety [21:59] the problem with the kvm flavor is that it's a wildly different beast [22:01] TJ-: oh you're right, i didn't read which message that was properly [22:02] TJ-: OK but if it's for safety purposes, shouldn't the kernel see it (and count it in MemTotal)? [22:05] "dmesg | grep BIOS-e820" shows 5 ranges as reserved [22:06] sdeziel: right those are bios reserved, and then the kernel removes a bit more [22:06] sdeziel: no, because those reserved areas are often 'ROM' - e.g. system BIOS and option ROMs, or PCI device mappings [22:08] hmm [22:08] sdeziel: see e.g. "sudo less /proc/iomem" [22:10] "000c0000-000c0dff : Video ROM" => the VM only has a serial console [22:11] sdeziel: but the VM seabios presumably has a default e820 reservation for a standard VGA framebuffer [22:14] TJ-: I don't know that area at all but I would think that such mapping wouldn't remove RAM visible for the kernel [22:19] In the dmesg Memory line you referenced, the 'reserved' is: (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10), [22:24] not sure where totalram_pages comes from though [22:26] I'm just lacking too much basic knowledge I guess I should start reading https://www.kernel.org/doc/Documentation/devicetree/bindings/reserved-memory/reserved-memory.txt [22:28] huh, this ^ not related at all [22:31] no, that's about device-tree bindings [22:33] TJ-: yeah, I noticed. While I look around for documentation of the e820 memory mapping hints, do you know how I could minimize the amount of reserved memory? [22:33] TJ-: something to tweak with QEMU/seabios? [22:34] sdeziel: did you pastebin the entire dmesg ? [22:35] TJ-: https://paste.ubuntu.com/p/8ysM3kGX4t/ [22:36] sdeziel: and what's in /proc/iomem [22:36] qemu as invoked by libvirt: https://paste.ubuntu.com/p/FX8RtzN5wK/ [22:37] TJ-: iomem: https://paste.ubuntu.com/p/NV45Nk9bjm/ [23:02] hehe [23:02] any new exciting news? [23:03] yes [23:04] cool [23:52] TJ-: nacc: thanks for the useful information [23:53] sdeziel: yw, hopefully you were able to figure stuff out? [23:53] nacc: not yet but I have some pointers now at least