Qboot experiment and initcall probe - #12

Merged

aledbf merged 18 commits into

mainfrom

qboot-experiment-and-initcall-probe

Sep 26, 2026

Commits on Sep 26, 2026

The qboot experiment: a pinned build, and a probe that asserts before it times

aledbf
and
claude
committed
21bfd95
boot: where the kernel's own boot goes, initcall by initcall

aledbf
and
claude
committed
1d56d76
boot: two probes for early userspace, and three ways it loses data

aledbf
and
claude
committed
301fe71
boot: what udev actually walks, and why there is no single win in it

aledbf
and
claude
committed
ef96390
boot: udev is not on the critical path, and three rankings said it was

aledbf
and
claude
committed
3819a79
boot: the initcall probe uses --profile instead of redefining it

aledbf
and
claude
committed
2e5b35a
boot: clarify that releases do not include software that owns a guest
aledbf
committed
4dc5053
kernel: a floor, and what this machine's kernel configuration actually costs

aledbf
and
claude
committed
e8c4c27
boot: keep the conclusions near the code, not the sample tables

aledbf
and
claude
committed
e0254c5
kernel: erofs goes; nothing has mounted one since qcow2 replaced that path

aledbf
and
claude
committed
f85ae64
boot: measure the console swap from inside, and eliminate a third candidate

aledbf
and
claude
committed
7af74bc
image: TERM=dumb on the console unit, which is worth 331 ms

aledbf
and
claude
committed
662a170
kernel: KSM and THP are not the remaining candidate either

aledbf
and
claude
committed
74dfd72
boot: the number of unit files is not the boot, and the caches are already warm

aledbf
and
claude
committed
c3582c1
boot: ask systemd what unit loading costs, and it is 26 ms not 212

aledbf
and
claude
committed
ea8e086
boot: disable unnecessary systemd generators and virtual console support for optimized VM performance
aledbf
committed
bfe9b17
kernel: CONFIG_VT off as dead configuration, and a number withdrawn

aledbf
and
claude
committed
19b6a5d
boot: the two largest items in QEMU's startup are cold-boot only

aledbf
and
claude
committed
d260061

Navigation Menu

Qboot experiment and initcall probe - #12

Qboot experiment and initcall probe#12
aledbf merged 18 commits into
mainfrom
qboot-experiment-and-initcall-probe

Commits on Sep 26, 2026

The qboot experiment: a pinned build, and a probe that asserts before it times

boot: where the kernel's own boot goes, initcall by initcall

boot: two probes for early userspace, and three ways it loses data

boot: what udev actually walks, and why there is no single win in it

boot: udev is not on the critical path, and three rankings said it was

boot: the initcall probe uses --profile instead of redefining it

boot: clarify that releases do not include software that owns a guest

kernel: a floor, and what this machine's kernel configuration actually costs

boot: keep the conclusions near the code, not the sample tables

kernel: erofs goes; nothing has mounted one since qcow2 replaced that path

boot: measure the console swap from inside, and eliminate a third candidate

image: TERM=dumb on the console unit, which is worth 331 ms

kernel: KSM and THP are not the remaining candidate either

boot: the number of unit files is not the boot, and the caches are already warm

boot: ask systemd what unit loading costs, and it is 26 ms not 212

boot: disable unnecessary systemd generators and virtual console support for optimized VM performance

kernel: CONFIG_VT off as dead configuration, and a number withdrawn

boot: the two largest items in QEMU's startup are cold-boot only