Control: severity -1 important Control: tag -1 moreinfo On Mon, 29 Jan 2018 15:05:00 +0000 Chris Boot <bootc@debian.org> wrote: > Package: src:linux > Followup-For: Bug #883413 > > Hi Ben, > > Unfortunately I can still reproduce this problem on 4.15-rc8 from > experimental. > > The cmdline for this boot was: > > BOOT_IMAGE=/boot/vmlinuz-4.15.0-rc8-amd64 > root=/dev/mapper/vg_tarquin-rootfs ro intel_iommu=on vsyscall=emulate > scsi_mod.use_blk_mq=Y dm_mod.use_blk_mq=Y intel_pstate=passive > i915.disable_display=Y i915.enable_gvt=Y apparmor=0 > systemd.unified_cgroup_hierarchy=1 console=ttyS1,115200n8 console=tty0 > > This triggers with DefaultMemoryAccounting=yes enabled in > /etc/systemd/system.conf, and NUT seems to regularly be involved in the > crash on my system. Sadly the systemd unit is very simple indeed, and > because my UPS is network-connected I'm not even doing dodgy things like > USB from within NUT. > > Quite how the kernel thinks that nut-server.service is using 16 ZiB of > memory is beyond me; presumably this is a slightly negative 64-bit int > bring cast unsigned. The following also feels like a smoking gun: > > [ 2982.158622] percpu ref (css_release) <= 0 (-197) after switching to atomic [...] Sorry for leaving this unanswered so long. Are you still seeing this? I found some apparently related reports on the Red Hat Bugzilla but not on anything newer than 4.17. Ben. -- Ben Hutchings The world is coming to an end. Please log off.
Attachment:
signature.asc
Description: This is a digitally signed message part