@GrahamcOfBorg build linuxPackages_5_0.zfs linuxPackages_5_1.zfs linuxPackages.zfs

one more try (timed out on x86_64)!

Ah yes, this needs #61181 as well for it to actually fix the spl build. #61181 is now merged

cc @thoughtpolice who seems to be the only maintainer listed for the kernel, @NeQuissimus who does a lot of kernel stuff

Hm, perhaps I'm missing something but on master I get

% grep PREEMPT $(nix-build -A linux_latest.configfile)
# CONFIG_PREEMPT_NONE is not set
CONFIG_PREEMPT_VOLUNTARY=y
# CONFIG_PREEMPT is not set

@joachifm it only seems to default to y on aarch64.

LGTM

Sorry about the miss here @lheckemann, thanks for the patch! LGTM.

hi folks, wondering about that 2d44588 commit added after the merge (use preempt voluntary only for aarch64)
If the zfs problem is isolated to aarch64, can we keep PREEMPT instead of voluntary for better default x86_64 realtime performance?

@digitalsignalperson the main reason for applying it only to aarch64 was to minimise the amount of changes to the extent that nothing even gets rebuilt on x86_64. I'd suggest setting CONFIG_PREEMPT_VOLUNTARY to lib.mkForce no for musnix, and stick to the upstream default in nixpkgs itself.

@lheckemann Sorry I think I'm confused. It looks to me that 1d37727 was merged into master which applies the change to both x86_64 and aarch64. Is that not the case?

in other words, on my x86_64 system on nixos-unstable, output from zgrep PREEMPT /proc/config.gz includes

CONFIG_PREEMPT_VOLUNTARY=y
# CONFIG_PREEMPT is not set

@digitalsignalperson yes, that is correct, however this was already the case before 1d37727, the only change being that it's now defined explicitly rather than coming from the kernel defconfig.

Ah, I see. Thanks for clarifying