Hello,
Mainline kernel 6.11.1 on my archlinux machine with several disks, if
I apply "none" as the io-scheduler, it core dumps.
Startup script rc.local has:
echo none | sudo tee /sys/block/nvme{0n1,1n1}/queue/scheduler
Causes this in the journal log -
Oct 01 14:39:06 tor systemd-coredump[987]: [🡕] Process 980 (tee) of
user 0 dumped core.
Stack trace of thread 980:
#0 0x000079b6a0ca43f4 n/a
(libc.so.6 + 0x963f4)
#1 0x000079b6a0c4b120
raise (libc.so.6 + 0x3d120)
#2 0x000079b6a0c324c3
abort (libc.so.6 + 0x244c3)
#3 0x000079b6a0c323df n/a
(libc.so.6 + 0x243df)
#4 0x000079b6a0c43177
__assert_fail (libc.so.6 + 0x35177)
#5 0x0000617a0fb4e3ec n/a
(tee + 0x63ec)
#6 0x0000617a0fb4a34f n/a
(tee + 0x234f)
#7 0x000079b6a0c33e08 n/a
(libc.so.6 + 0x25e08)
#8 0x000079b6a0c33ecc
__libc_start_main (libc.so.6 + 0x25ecc)
#9 0x0000617a0fb4a665 n/a
(tee + 0x2665)
ELF object binary
architecture: AMD x86-64
Problem is reproducible as well.
I can set the nvme disks for "bfq" but then cannot set it back to "none".
Unsure about the origin of the problem but it appears to be something
with libc and the newer kernel.
I can login as root, and redirect ( echo none >
/proc/sys/nvme1n1/queue/scheduler ... ) this procuded no error but the
scheduler does not change.
Also tried on a Rocky 9.4 VM with the mainline 6.11.1 elrepo kernel,
the error just reports no such file or directory ... the VM also has
virtio devices so unsure if that is related.
I can provide other outputs of commands like strace if needed.
-- Fred R.
