| 10 Mar 2026 |
 amaan | yeah true;
let
pageSizeMap = {
"4" = 12;
"16" = 14;
"64" = 16;
};
in
assert lib.asserts.assertMsg (builtins.hasAttr (toString pageSizeKiB) pageSizeMap)
"jemalloc: pageSizeKiB must be one of ${lib.concatStringsSep ", " (builtins.attrNames pageSizeMap)}, but is: ${toString pageSizeKiB}";
is a bit better then | 06:38:14 |
 Puna | i think at that point you're just reinventing assertOneOf, but without the multi-line printing 😅 | 06:39:14 |
| amaan | true...then that's fine | 06:39:24 |
| amaan | (i do like the presentation a little bit more but it's nbd) | 06:40:02 |
| amaan | ok, just pushed that | 06:41:54 |
| Puna | seems fine to me, but it's not my package. maybe someone else here who has an opinion on exposing jemalloc's page size setting like this?
https://github.com/NixOS/nixpkgs/pull/498442 | 07:54:27 |
 Alyssa Ross | IMO we should always use the highest possible page size for jemalloc | 11:08:36 |
| Alyssa Ross | so this is a good change | 11:08:41 |
| Puna | kernel apparently also has a setting for 256k page size, CONFIG_PPC_256K_PAGES for POWER. says it needs patched binutils though. | 11:27:53 |
| Puna | not something i wanna test 😅 | 11:28:07 |
|  Aelin joined the room. | 16:11:19 |
| Aelin | I've never seen a 256k page size POWER system, 64k is the standard nowadays so that change is very reasonable | 16:12:34 |
| Aelin | so, ppc64 VMs are special because they basically scan all disks (which should use MBR, but SLOF can also handle GPT) for a ppc prep boot partition, which is the partition that GRUB is installed to | 16:14:24 |
| Aelin | the prep boot partition needs to be marked as bootable | 16:14:34 |
| Aelin | this is entirely unlike any physical ppc64 hardware which in OPAL just all use petitboot to my knowledge | 16:15:05 |
| 11 Mar 2026 |
|  @somasis:matrix.org left the room. | 03:49:38 |
 Stas | 
Download 1000005234.jpg | 19:31:35 |
| Stas | Burn in test in progress | 19:31:54 |
| Stas | Once complete, I will finally get to build some nixos | 19:32:10 |
| Stas | One of the ram sticks was slotted in badly | 19:33:27 |
| Stas | Reseating got my 16g back | 19:33:42 |
| Puna | 4x as much real RAM as i have, so building everything might be alot faster for you 😅 | 19:35:44 |
| Puna | 
Download image.png | 19:36:02 |
| Stas | :D | 19:36:41 |
| Puna | (add ~ 1d or so to that elapsed build time, hit an openssl regression p early that required a PR) | 19:37:13 |
| Stas | I have 2 chinese sticks left and original ones. Can ship them to you | 19:37:42 |
| Aelin | if it helps any of you for development I can spin up some more ppc64 VMs on POWER8 hw like I did for amaan | 19:38:20 |
| Stas |
https://a.aliexpress.com/_EQNbDhE | 19:38:21 |
| Stas | Got 10 of these from ali | 19:38:31 |
| Puna | might come back to you on that when i have less other things that i need to do. the slow pace of these builds helps me w/ not putting too many hours at a time into this ^^" | 19:46:18 |
